def main(args):
with tf.Graph().as_default():
with tf.Session() as sess:
# Read the file containing the pairs used for testing
pairs = lfw.read_pairs(os.path.expanduser(args.lfw_pairs))
print('')
print('pair: ', pairs)
print('')
print('args:', os.path.expanduser(args.lfw_pairs))
print('')
# Get the paths for the corresponding images
paths, actual_issame = lfw.get_paths(os.path.expanduser(args.lfw_dir), pairs)
image_paths_placeholder = tf.placeholder(tf.string, shape=(None,1), name='image_paths')
labels_placeholder = tf.placeholder(tf.int32, shape=(None,1), name='labels')
batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size')
control_placeholder = tf.placeholder(tf.int32, shape=(None,1), name='control')
phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train')
nrof_preprocess_threads = 4
image_size = (args.image_size, args.image_size)
eval_input_queue = data_flow_ops.FIFOQueue(capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1,), (1,), (1,)],
shared_name=None, name=None)
eval_enqueue_op = eval_input_queue.enqueue_many([image_paths_placeholder, labels_placeholder, control_placeholder], name='eval_enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(eval_input_queue, image_size, nrof_preprocess_threads, batch_size_placeholder)
# Load the model
input_map = {'image_batch': image_batch, 'label_batch': label_batch, 'phase_train': phase_train_placeholder}
facenet.load_model(args.model, input_map=input_map)
# Get output tensor
embeddings = tf.get_default_graph().get_tensor_by_name("embeddings:0")
#
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
evaluate(sess,
eval_enqueue_op,
image_paths_placeholder,
labels_placeholder,
phase_train_placeholder,
batch_size_placeholder,
control_placeholder,
embeddings,
label_batch,
paths,
actual_issame,
args.lfw_batch_size,
args.lfw_nrof_folds,
args.distance_metric,
args.subtract_mean,
args.use_flipped_images,
args.use_fixed_image_standardization)
def main(args):
with tf.Graph().as_default():
with tf.Session() as sess:
# Read the file containing the pairs used for testing
pairs = lfw.read_pairs(os.path.expanduser(args.lfw_pairs))
# Get the paths for the corresponding images
paths, actual_issame = lfw.get_paths(os.path.expanduser(args.lfw_dir), pairs)
image_paths_placeholder = tf.placeholder(tf.string, shape=(None,1), name='image_paths')
labels_placeholder = tf.placeholder(tf.int32, shape=(None,1), name='labels')
batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size')
control_placeholder = tf.placeholder(tf.int32, shape=(None,1), name='control')
phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train')
nrof_preprocess_threads = 4
image_size = (args.image_size, args.image_size)
eval_input_queue = data_flow_ops.FIFOQueue(capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1,), (1,), (1,)],
shared_name=None, name=None)
eval_enqueue_op = eval_input_queue.enqueue_many([image_paths_placeholder, labels_placeholder, control_placeholder], name='eval_enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(eval_input_queue, image_size, nrof_preprocess_threads, batch_size_placeholder)
# Load the model
input_map = {'image_batch': image_batch, 'label_batch': label_batch, 'phase_train': phase_train_placeholder}
facenet.load_model(args.model, input_map=input_map)
# Get output tensor
embeddings = tf.get_default_graph().get_tensor_by_name("embeddings:0")
#
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
evaluate(sess, eval_enqueue_op, image_paths_placeholder, labels_placeholder, phase_train_placeholder, batch_size_placeholder, control_placeholder,
embeddings, label_batch, paths, actual_issame, args.lfw_batch_size, args.lfw_nrof_folds, args.distance_metric, args.subtract_mean,
args.use_flipped_images, args.use_fixed_image_standardization)
def main():
with tf.Graph().as_default():
with tf.Session() as sess:
# Read the file containing the pairs used for testing
pairs = read_pairs(os.path.expanduser(ACD_PAIRS))
# Get the paths for the corresponding images
paths, actual_issame = get_paths(os.path.expanduser(ACD_DIR), pairs)
image_paths_placeholder = tf.placeholder(tf.string, shape=(None,1), name='image_paths')
labels_placeholder = tf.placeholder(tf.int32, shape=(None,1), name='labels')
batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size')
control_placeholder = tf.placeholder(tf.int32, shape=(None,1), name='control')
phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train')
nrof_preprocess_threads = 4
image_size = (IMAGE_SIZE, IMAGE_SIZE)
eval_input_queue = data_flow_ops.FIFOQueue(capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1,), (1,), (1,)],
shared_name=None, name=None)
eval_enqueue_op = eval_input_queue.enqueue_many([image_paths_placeholder, labels_placeholder, control_placeholder], name='eval_enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(eval_input_queue, image_size, nrof_preprocess_threads, batch_size_placeholder)
# Load the model
input_map = {'image_batch': image_batch, 'label_batch': label_batch, 'phase_train': phase_train_placeholder}
facenet.load_model(MODEL_DIR, input_map=input_map)
# Get output tensor
embeddings = tf.get_default_graph().get_tensor_by_name("embeddings:0")
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
evaluate_tsdvface(sess, eval_enqueue_op, image_paths_placeholder, labels_placeholder, phase_train_placeholder, batch_size_placeholder, control_placeholder,
embeddings, label_batch, paths, actual_issame, ACD_BATCH_SIZE, ACD_NROF_FOLDS, DISTANCE_METRIC, SUBTRACT_MEAN,
USE_FLIPPED_IMAGES, USE_FIXED_STD)
def get_similarity(config_, input_path_a, input_path_b, multiple_pairs=False):
config_.display()
# input image size, can be changed in config.py
image_size = (config_.im_height, config_.im_width)
if not multiple_pairs:
# preprocess the input images
preprocess(config_, input_path_a, input_path_b)
# get the updated values of image names
tf.reset_default_graph()
image_path_a, ext_a = os.path.splitext(os.path.split(input_path_a)[1])
image_path_b, ext_b = os.path.splitext(os.path.split(input_path_b)[1])
image_path_a = os.path.join(config_.preprocessed_out_dir_test,
image_path_a + '_cropped' + ext_a)
image_path_b = os.path.join(config_.preprocessed_out_dir_test,
image_path_b + '_cropped' + ext_b)
# create the paths for inference
paths, actual_issame = [(image_path_a, image_path_b),
(image_path_a, image_path_b),
(image_path_a, image_path_b)], [False] * 3
if multiple_pairs:
input_path_a_list = list(pd.read_csv(input_path_a)["paths"])
input_path_b_list = list(pd.read_csv(input_path_b)["paths"])
preprocess(config_, input_path_a, input_path_b, 'true')
paths, actual_issame = [], []
for i in range(len(input_path_a_list)):
input_path_a, input_path_b = input_path_a_list[
i], input_path_b_list[i]
image_path_a, ext_a = os.path.splitext(
os.path.split(input_path_a)[1])
image_path_b, ext_b = os.path.splitext(
os.path.split(input_path_b)[1])
image_path_a = os.path.join(config_.preprocessed_out_dir_test,
image_path_a + '_cropped' + ext_a)
image_path_b = os.path.join(config_.preprocessed_out_dir_test,
image_path_b + '_cropped' + ext_b)
paths.append((image_path_a, image_path_b))
actual_issame.append(False) # dummy values
with tf.Graph().as_default():
with tf.Session() as sess:
image_paths_placeholder = tf.placeholder(tf.string,
shape=(None, 1),
name='image_paths')
labels_placeholder = tf.placeholder(
tf.int32, shape=(None, 1),
name='labels') # labels of the input pairs
batch_size_placeholder = tf.placeholder(
tf.int32,
name='batch_size') # batch size for evaluation, default =1
control_placeholder = tf.placeholder(
tf.int32, shape=(None, 1), name='control'
) # control argument for data augmentation, not needed for testing
phase_train_placeholder = tf.placeholder(
tf.bool, name='phase_train') # flag train(True)/test(False)
nrof_preprocess_threads = 4
eval_input_queue = data_flow_ops.FIFOQueue(
capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1, ), (1, ), (1, )],
shared_name=None,
name=None)
eval_enqueue_op = eval_input_queue.enqueue_many(
[
image_paths_placeholder, labels_placeholder,
control_placeholder
],
name='eval_enqueue_op')
# input pipeline, needed to define the input/output signatures
image_batch, label_batch = facenet.create_input_pipeline(
eval_input_queue, image_size, nrof_preprocess_threads,
batch_size_placeholder)
# Load the model, modify the path to pretrained model in config.py
input_map = {
'image_batch': image_batch,
'label_batch': label_batch,
'phase_train': phase_train_placeholder
}
facenet.load_model(config_.pretrained_model, input_map=input_map)
# Get output tensor, needed to create inference graph
embeddings = tf.get_default_graph().get_tensor_by_name(
"embeddings:0")
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
similarity = evaluate(sess, eval_enqueue_op,
image_paths_placeholder, labels_placeholder,
phase_train_placeholder,
batch_size_placeholder, control_placeholder,
embeddings, label_batch, paths,
actual_issame, config_.pred_batch_size)
# write the output to json
# similarity threshold calculated based on validation performance
are_same = ["Yes" if dist < 0.38 else "No" for dist in similarity]
similarity = [str(dist) for dist in similarity]
out_data = {
"similarity_score": similarity,
"are_same_people": are_same
}
#save the info to json file
return similarity, out_data
def main(args):
with tf.Graph().as_default():
with tf.Session() as sess:
# Read the file containing the pairs used for testing
# 读取包含用于 test 的 pairs 的文件
pairs = lfw.read_pairs(os.path.expanduser(args.lfw_pairs))
#pdb.set_trace()
# Get the paths for the corresponding images
# 获取相应图像的路径
paths, actual_issame = lfw.get_paths(
os.path.expanduser(args.lfw_dir), pairs, args.lfw_file_ext)
network = importlib.import_module(args.model_def)
image_paths_placeholder = tf.placeholder(tf.string,
shape=(None, 1),
name='image_paths')
labels_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='labels')
batch_size_placeholder = tf.placeholder(tf.int32,
name='batch_size')
control_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='control')
phase_train_placeholder = tf.placeholder(tf.bool,
name='phase_train')
nrof_preprocess_threads = 4
image_size = (args.image_size, args.image_size)
eval_input_queue = data_flow_ops.FIFOQueue(
capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1, ), (1, ), (1, )],
shared_name=None,
name=None)
eval_enqueue_op = eval_input_queue.enqueue_many(
[
image_paths_placeholder, labels_placeholder,
control_placeholder
],
name='eval_enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(
eval_input_queue, image_size, nrof_preprocess_threads,
batch_size_placeholder)
# Load the model
input_map = {
'image_batch': image_batch,
'label_batch': label_batch,
'phase_train': phase_train_placeholder
}
facenet.load_model(args.model, input_map=input_map)
prelogits, _ = network.inference(
image_batch,
args.keep_probability,
phase_train=phase_train_placeholder,
bottleneck_layer_size=args.embedding_size,
weight_decay=args.weight_decay)
embeddings = tf.nn.l2_normalize(prelogits,
1,
1e-10,
name='embeddings')
# embeddings = tf.identity(prelogits)
# Initialize variables
sess.run(tf.global_variables_initializer(),
feed_dict={phase_train_placeholder: False})
sess.run(tf.local_variables_initializer(),
feed_dict={phase_train_placeholder: False})
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
evaluate(sess, eval_enqueue_op, image_paths_placeholder,
labels_placeholder, phase_train_placeholder,
batch_size_placeholder, control_placeholder, embeddings,
label_batch, paths, actual_issame, args.lfw_batch_size,
args.lfw_nrof_folds, args.distance_metric,
args.subtract_mean, args.use_flipped_images,
args.use_fixed_image_standardization)
def main(args):
# args从文件系统中导入对象
network = importlib.import_module(
args.model_def) # 导入模型，default = 'models.inception_resnet_v1'
image_size = (args.image_size, args.image_size) # 导入尺寸，default = 160
subdir = datetime.strftime(datetime.now(), '%Y%m%d-%H%M%S')
log_dir = os.path.join(os.path.expanduser(args.logs_base_dir),
subdir) # 导入运行日志目录，default = '~/logs/facenet'
if not os.path.isdir(
log_dir): # Create the log directory if it doesn't exist
os.makedirs(log_dir)
model_dir = os.path.join(os.path.expanduser(args.models_base_dir),
subdir) # ，导入模型目录，default = '~/models/facenet'
if not os.path.isdir(
model_dir): # Create the model directory if it doesn't exist
os.makedirs(model_dir)
stat_file_name = os.path.join(log_dir, 'stat.h5') # 统计数据，用h5格式保存
# Write arguments to a text file
facenet.write_arguments_to_file(args,
os.path.join(log_dir,
'arguments.txt')) # 记录命令至运行日志
# Store some git revision info in a text file in the log directory
src_path, _ = os.path.split(
os.path.realpath(__file__)) # os.path.realpath(__file__)是脚本所在的绝对路径
facenet.store_revision_info(src_path, log_dir, ' '.join(sys.argv))
np.random.seed(seed=args.seed) # default = 666
random.seed(args.seed)
dataset = facenet.get_dataset(
args.data_dir
) # 导入数据集目录，default = '~/datasets/casia/casia_maxpy_mtcnnalign_182_160'
if args.filter_filename: # default = ''
dataset = filter_dataset(
dataset,
os.path.expanduser(args.filter_filename),
# expanduser把path中包含的"~"和"~user"转换成用户目录.
args.filter_percentile,
args.filter_min_nrof_images_per_class
) # default = 100.0, default = 0
if args.validation_set_split_ratio > 0.0: # default = 0.0
train_set, val_set = facenet.split_dataset(
dataset, args.validation_set_split_ratio,
args.min_nrof_val_images_per_class, 'SPLIT_IMAGES')
else:
train_set, val_set = dataset, []
nrof_classes = len(train_set)
print('Model directory: %s' % model_dir)
print('Log directory: %s' % log_dir)
pretrained_model = None
if args.pretrained_model:
pretrained_model = os.path.expanduser(args.pretrained_model)
print('Pre-trained model: %s' % pretrained_model)
if args.lfw_dir:
print('LFW directory: %s' % args.lfw_dir)
# Read the file containing the pairs used for testing
pairs = lfw.read_pairs(os.path.expanduser(
args.lfw_pairs)) # default = 'data/pairs.txt'
# Get the paths for the corresponding images
lfw_paths, actual_issame = lfw.get_paths(
os.path.expanduser(args.lfw_dir), pairs)
with tf.Graph().as_default():
tf.set_random_seed(args.seed) # 图级随机数种子
global_step = tf.Variable(0, trainable=False)
# Get a list of image paths and their labels
image_list, label_list = facenet.get_image_paths_and_labels(train_set)
assert len(image_list) > 0, 'The training set should not be empty'
val_image_list, val_label_list = facenet.get_image_paths_and_labels(
val_set)
# Create a queue that produces indices into the image_list and label_list
labels = ops.convert_to_tensor(label_list, dtype=tf.int32)
range_size = array_ops.shape(labels)[0]
index_queue = tf.train.range_input_producer(
range_size,
num_epochs=None, # 产生1到range_size的对列，num_epochs不给值的话
shuffle=True,
seed=None,
capacity=32) # 则默认无限循环，后面用capacity限制容量
index_dequeue_op = index_queue.dequeue_many(
args.batch_size * args.epoch_size, 'index_dequeue') # 使多个元素同时出列并命名
# 创建张量
learning_rate_placeholder = tf.placeholder(tf.float32,
name='learning_rate')
batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size')
phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train')
image_paths_placeholder = tf.placeholder(tf.string,
shape=(None, 1),
name='image_paths')
labels_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='labels')
control_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='control')
nrof_preprocess_threads = 4
# 先进先出对列
input_queue = data_flow_ops.FIFOQueue(
capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1, ), (1, ), (1, )],
shared_name=None,
name=None)
enqueue_op = input_queue.enqueue_many(
[image_paths_placeholder, labels_placeholder, control_placeholder],
name='enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(
input_queue, image_size, nrof_preprocess_threads,
batch_size_placeholder)
image_batch = tf.identity(image_batch, 'image_batch')
image_batch = tf.identity(image_batch, 'input')
label_batch = tf.identity(label_batch, 'label_batch')
print('Number of classes in training set: %d' % nrof_classes)
print('Number of examples in training set: %d' % len(image_list))
print('Number of classes in validation set: %d' % len(val_set))
print('Number of examples in validation set: %d' % len(val_image_list))
print('Building training graph')
# Build the inference graph
prelogits, _ = network.inference(
image_batch,
args.keep_probability,
phase_train=phase_train_placeholder,
bottleneck_layer_size=args.embedding_size,
weight_decay=args.weight_decay)
logits = slim.fully_connected(
prelogits,
len(train_set),
activation_fn=None, # 相当于Dense
weights_initializer=slim.initializers.xavier_initializer(),
weights_regularizer=slim.l2_regularizer(args.weight_decay),
scope='Logits',
reuse=False)
embeddings = tf.nn.l2_normalize(
prelogits, 1, 1e-10, name='embeddings') # 按行进行L2正则化，同除各行L2范数
# Norm for the prelogits
eps = 1e-4
# 按行计算平均值。norm计算范数，abs取绝对值，ord取1，使用第一范数
prelogits_norm = tf.reduce_mean(
tf.norm(tf.abs(prelogits) + eps, ord=args.prelogits_norm_p,
axis=1))
tf.add_to_collection(
tf.GraphKeys.REGULARIZATION_LOSSES, prelogits_norm *
args.prelogits_norm_loss_factor) # 将该种正则化方式储存至graphkeys
# Add center loss
prelogits_center_loss, _ = facenet.center_loss(prelogits, label_batch,
args.center_loss_alfa,
nrof_classes)
tf.add_to_collection(tf.GraphKeys.REGULARIZATION_LOSSES,
prelogits_center_loss * args.center_loss_factor)
# 指数衰减。参数1：原始学习率。参数2：全局步数，每次自增1，如果不写则不增1.参数3：衰减系数。参数4：衰减速度。参数5：是否取整。True时取整
# 具体计算：decayed_learning_rate = learning_rate*decay_rate^(global_step/decay_steps)
learning_rate = tf.train.exponential_decay(
learning_rate_placeholder,
global_step,
args.learning_rate_decay_epochs * args.epoch_size,
args.learning_rate_decay_factor,
staircase=True)
tf.summary.scalar('learning_rate', learning_rate) # 保存数据
# Calculate the average cross entropy loss across the batch
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(
labels=label_batch,
logits=logits,
name='cross_entropy_per_example')
cross_entropy_mean = tf.reduce_mean(cross_entropy,
name='cross_entropy')
tf.add_to_collection('losses', cross_entropy_mean)
# cast转换数据类型(布尔->浮点数)，equal比较两个向量相同位置的元素是否一样
correct_prediction = tf.cast(
tf.equal(tf.argmax(logits, 1), tf.cast(label_batch, tf.int64)),
tf.float32)
accuracy = tf.reduce_mean(correct_prediction) # 求平均值即准确率
# Calculate the total losses
regularization_losses = tf.get_collection(
tf.GraphKeys.REGULARIZATION_LOSSES) # 从graphkeys中把之前存储的loss计算方式取出来
total_loss = tf.add_n([cross_entropy_mean] + regularization_losses,
name='total_loss')
# Build a Graph that trains the model with one batch of examples and updates the model parameters
train_op = facenet.train(total_loss, global_step, args.optimizer,
learning_rate, args.moving_average_decay,
tf.global_variables(), args.log_histograms)
# Create a saver
saver = tf.train.Saver(tf.trainable_variables(), max_to_keep=3)
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.summary.merge_all(
) # 将所有summary全部保存到磁盘，以便tensorboard显示
# Start running operations on the Graph.
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=args.
gpu_memory_fraction) # 使用gpu，参数为使用的显存上限
sess = tf.Session(config=tf.ConfigProto(
gpu_options=gpu_options, log_device_placement=False)) # 会话
sess.run(tf.global_variables_initializer()) # 返回所有全局变量
sess.run(tf.local_variables_initializer()) # 返回所有局部变量
summary_writer = tf.summary.FileWriter(
log_dir, sess.graph) # 将训练过程数据保存在filewriter指定的文件中
coord = tf.train.Coordinator() # 管理在Session中的多个线程
"""
启动入队线程，由多个或单个线程，按照设定规则，把文件读入Filename Queue中。函数返回线程ID的列表，一般情况下，系统有多少个核，
就会启动多少个入队线程（入队具体使用多少个线程在tf.train.batch中定义）
"""
tf.train.start_queue_runners(coord=coord, sess=sess)
with sess.as_default():
if pretrained_model:
print('Restoring pretrained model: %s' % pretrained_model)
saver.restore(sess, pretrained_model)
# Training and validation loop
print('Running training')
nrof_steps = args.max_nrof_epochs * args.epoch_size
nrof_val_samples = int(
math.ceil(args.max_nrof_epochs / args.validate_every_n_epochs)
) # Validate every validate_every_n_epochs as well as in the last epoch
# ceil返回上入整数
stat = {
'loss':
np.zeros((nrof_steps, ), np.float32),
'center_loss':
np.zeros((nrof_steps, ), np.float32),
'reg_loss':
np.zeros((nrof_steps, ), np.float32),
'xent_loss':
np.zeros((nrof_steps, ), np.float32),
'prelogits_norm':
np.zeros((nrof_steps, ), np.float32),
'accuracy':
np.zeros((nrof_steps, ), np.float32),
'val_loss':
np.zeros((nrof_val_samples, ), np.float32),
'val_xent_loss':
np.zeros((nrof_val_samples, ), np.float32),
'val_accuracy':
np.zeros((nrof_val_samples, ), np.float32),
'lfw_accuracy':
np.zeros((args.max_nrof_epochs, ), np.float32),
'lfw_valrate':
np.zeros((args.max_nrof_epochs, ), np.float32),
'learning_rate':
np.zeros((args.max_nrof_epochs, ), np.float32),
'time_train':
np.zeros((args.max_nrof_epochs, ), np.float32),
'time_validate':
np.zeros((args.max_nrof_epochs, ), np.float32),
'time_evaluate':
np.zeros((args.max_nrof_epochs, ), np.float32),
'prelogits_hist':
np.zeros((args.max_nrof_epochs, 1000), np.float32),
}
for epoch in range(1, args.max_nrof_epochs + 1):
step = sess.run(global_step, feed_dict=None)
# Train for one epoch
t = time.time()
cont = train(
args, sess, epoch, image_list, label_list,
index_dequeue_op, enqueue_op, image_paths_placeholder,
labels_placeholder, learning_rate_placeholder,
phase_train_placeholder, batch_size_placeholder,
control_placeholder, global_step, total_loss, train_op,
summary_op, summary_writer, regularization_losses,
args.learning_rate_schedule_file, stat, cross_entropy_mean,
accuracy, learning_rate, prelogits, prelogits_center_loss,
args.random_rotate, args.random_crop, args.random_flip,
prelogits_norm, args.prelogits_hist_max,
args.use_fixed_image_standardization)
stat['time_train'][epoch - 1] = time.time() - t
if not cont:
break
t = time.time()
if len(val_image_list) > 0 and (
(epoch - 1) % args.validate_every_n_epochs
== args.validate_every_n_epochs - 1
or epoch == args.max_nrof_epochs):
validate(args, sess, epoch, val_image_list, val_label_list,
enqueue_op, image_paths_placeholder,
labels_placeholder, control_placeholder,
phase_train_placeholder, batch_size_placeholder,
stat, total_loss, regularization_losses,
cross_entropy_mean, accuracy,
args.validate_every_n_epochs,
args.use_fixed_image_standardization)
stat['time_validate'][epoch - 1] = time.time() - t
# Save variables and the metagraph if it doesn't exist already
save_variables_and_metagraph(sess, saver, summary_writer,
model_dir, subdir, epoch)
# Evaluate on LFW
t = time.time()
if args.lfw_dir:
evaluate(sess, enqueue_op, image_paths_placeholder,
labels_placeholder, phase_train_placeholder,
batch_size_placeholder, control_placeholder,
embeddings, label_batch, lfw_paths, actual_issame,
args.lfw_batch_size, args.lfw_nrof_folds, log_dir,
step, summary_writer, stat, epoch,
args.lfw_distance_metric, args.lfw_subtract_mean,
args.lfw_use_flipped_images,
args.use_fixed_image_standardization)
stat['time_evaluate'][epoch - 1] = time.time() - t
print('Saving statistics')
with h5py.File(stat_file_name, 'w') as f:
for key, value in stat.iteritems():
f.create_dataset(key, data=value)
return model_dir
def main(args):
with tf.Graph().as_default():
with tf.Session(config=config) as sess:
# Read the file containing the pairs used for testing
pairs = lfw.read_pairs(os.path.expanduser(args.lfw_pairs))
# Get the paths for the corresponding images
paths, actual_issame = lfw.get_paths(
os.path.expanduser(args.lfw_dir), pairs)
# add path from check_list
if args.check_list is not 'None':
print('Reading check_list')
paths, actual_issame = check.get_paths(args.lfw_dir,
args.check_list, paths,
actual_issame)
# check query_dir, this directory can only contain one query image
if args.query_dir is not 'None':
paths, actual_issame = check.get_query_path(
args.query_dir, paths, actual_issame)
print('{} calculation of embeddings remain'.format(paths.shape[0]))
if paths.shape[0] == 0:
sys.exit()
image_paths_placeholder = tf.placeholder(tf.string,
shape=(None, 1),
name='image_paths')
labels_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='labels')
batch_size_placeholder = tf.placeholder(tf.int32,
name='batch_size')
control_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='control')
phase_train_placeholder = tf.placeholder(tf.bool,
name='phase_train')
nrof_preprocess_threads = 4
image_size = (args.image_size, args.image_size)
eval_input_queue = data_flow_ops.FIFOQueue(
capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1, ), (1, ), (1, )],
shared_name=None,
name=None)
eval_enqueue_op = eval_input_queue.enqueue_many(
[
image_paths_placeholder, labels_placeholder,
control_placeholder
],
name='eval_enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(
eval_input_queue, image_size, nrof_preprocess_threads,
batch_size_placeholder)
# Load the model
input_map = {
'image_batch': image_batch,
'label_batch': label_batch,
'phase_train': phase_train_placeholder
}
facenet.load_model(args.model, input_map=input_map)
# Get output tensor
embeddings = tf.get_default_graph().get_tensor_by_name(
"embeddings:0")
#
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
# get the best batch size
batch_size = check.get_batch_size(args.lfw_batch_size,
paths.shape[0])
print('best batch_size: %d' % batch_size)
embeddings = evaluate(
sess, eval_enqueue_op, image_paths_placeholder,
labels_placeholder, phase_train_placeholder,
batch_size_placeholder, control_placeholder, embeddings,
label_batch, paths, actual_issame, batch_size,
args.lfw_nrof_folds, args.distance_metric, args.subtract_mean,
args.use_flipped_images, args.use_fixed_image_standardization)
check.save_embed(paths, embeddings)
def main(args):
network = importlib.import_module(args.model_def)
image_size = (args.image_size, args.image_size)
subdir = datetime.strftime(datetime.now(), '%Y%m%d-%H%M%S')
log_dir = os.path.join(os.path.expanduser(args.logs_base_dir),
subdir) #日志保存地址
if not os.path.isdir(
log_dir): # Create the log directory if it doesn't exist
os.makedirs(log_dir)
model_dir = os.path.join(os.path.expanduser(args.models_base_dir), subdir)
if not os.path.isdir(
model_dir): # Create the model directory if it doesn't exist
os.makedirs(model_dir)
# Write arguments to a text file
facenet.write_arguments_to_file(args, os.path.join(log_dir,
'arguments.txt'))
# Store some git revision info in a text file in the log directory
# os.path.realpath(__file__)代表返回当前模块真实路径
# os.path.split(os.path.realpath(__file__))代表返回路径的目录和文件名(元组形式)
src_path, _ = os.path.split(os.path.realpath(__file__))
facenet.store_revision_info(src_path, log_dir, ' '.join(sys.argv))
np.random.seed(seed=args.seed)
random.seed(args.seed)
# dataset是列表，列表元素为每一个类的ImageClass对象，定义见facenet.py文件
dataset = facenet.get_dataset(args.data_dir)
if args.filter_filename:
dataset = filter_dataset(dataset,
os.path.expanduser(args.filter_filename),
args.filter_percentile,
args.filter_min_nrof_images_per_class)
# 划分训练集和测试集，train_set和val_set的形式和dataset一样
if args.validation_set_split_ratio > 0.0:
train_set, val_set = facenet.split_dataset(
dataset, args.validation_set_split_ratio,
args.min_nrof_val_images_per_class, 'SPLIT_IMAGES')
else:
train_set, val_set = dataset, []
nrof_classes = len(train_set) #类目数(即有多少人)
print('Model directory: %s' % model_dir)
print('Log directory: %s' % log_dir)
pretrained_model = None
if args.pretrained_model:
pretrained_model = os.path.expanduser(args.pretrained_model)
print('Pre-trained model: %s' % pretrained_model)
if args.lfw_dir:
print('LFW directory: %s' % args.lfw_dir)
# Read the file containing the pairs used for testing
# pairs也是列表，子元素也是列表，每个子列表包含pairs.txt的每一行
pairs = lfw.read_pairs(os.path.expanduser(args.lfw_pairs))
# Get the paths for the corresponding images
#lfw_paths为两两比较的列表,列表的元素为有2个元素的元祖,actual_issame为列表,表示每个元素是否为同一个人
lfw_paths, actual_issame = lfw.get_paths(
os.path.expanduser(args.lfw_dir), pairs)
with tf.Graph().as_default():
tf.set_random_seed(args.seed)
# global_step对应的是全局批的个数，根据这个参数可以更新学习率
global_step = tf.Variable(0, trainable=False)
# Get a list of image paths and their labels
image_list, label_list = facenet.get_image_paths_and_labels(train_set)
assert len(image_list) > 0, 'The training set should not be empty'
val_image_list, val_label_list = facenet.get_image_paths_and_labels(
val_set)
# Create a queue that produces indices into the image_list and label_list
labels = ops.convert_to_tensor(label_list, dtype=tf.int32)
# range_size的大小和样本个数一样
range_size = array_ops.shape(labels)[0]
#QueueRunner：保存的是队列中的入列操作，保存在一个list当中，其中每个enqueue运行在一个线程当中
#range_input_producer：返回的是一个队列，队列中有打乱的整数，范围是从0到range size
#range_size大小和总的样本个数一样
#并将一个QueueRunner添加到当前图的QUEUE_RUNNER集合中
index_queue = tf.train.range_input_producer(
range_size, num_epochs=None, shuffle=True, seed=None,
capacity=32) #capacity代表队列容量
# 返回的是一个出列操作，每次出列一个epoch需要用到的样本个数
index_dequeue_op = index_queue.dequeue_many(
args.batch_size * args.epoch_size, 'index_dequeue')
learning_rate_placeholder = tf.placeholder(tf.float32,
name='learning_rate')
batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size')
phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train')
image_paths_placeholder = tf.placeholder(tf.string,
shape=(None, 1),
name='image_paths')
labels_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='labels')
control_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='control')
# 上面的队列是一个样本索引的出队队列，只用来出列
# 用来每次出列一个epoch中用到的样本
# 这里是第二个队列，这个队列用来入列，每个元素的大小为shape=[(1,),(1,),(1,)]
nrof_preprocess_threads = 4
# 这里的shape代表每一个元素的维度
input_queue = data_flow_ops.FIFOQueue(
capacity=600000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1, ), (1, ), (1, )],
shared_name=None,
name=None)
# 这时一个入列的操作，这个操作将在session run的时候用到
# 每次入列的是image_paths_placeholder, labels_placeholder对
# 注意，这里只有2个队列，一个用来出列打乱的元素序号
# 一个根据对应的需要读取指定的文件
enqueue_op = input_queue.enqueue_many(
[image_paths_placeholder, labels_placeholder, control_placeholder],
name='enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(
input_queue, image_size, nrof_preprocess_threads,
batch_size_placeholder)
# image_batch = tf.identity(image_batch, 'image_batch')
image_batch = tf.identity(image_batch, 'input')
label_batch = tf.identity(label_batch, 'label_batch')
print('Total number of classes: %d' % nrof_classes)
print('Total number of examples: %d' % len(image_list))
print('Building training graph')
# Build the inference graph
prelogits, _ = network.inference(
image_batch,
args.keep_probability,
phase_train=phase_train_placeholder,
bottleneck_layer_size=args.embedding_size,
weight_decay=args.weight_decay)
embeddings = tf.nn.l2_normalize(prelogits, 1, 1e-10, name='embeddings')
# arcface_logits = arcface_logits_compute(embeddings, label_batch, args, nrof_classes)
learning_rate = tf.train.exponential_decay(
learning_rate_placeholder,
global_step,
args.learning_rate_decay_epochs * args.epoch_size,
args.learning_rate_decay_factor,
staircase=True)
tf.summary.scalar('learning_rate', learning_rate)
with tf.variable_scope('Logits'):
cosface_logits = AM_logits_compute(embeddings, label_batch, args,
nrof_classes)
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(
labels=label_batch,
logits=cosface_logits,
name='cross_entropy_per_example')
cross_entropy_mean = tf.reduce_mean(cross_entropy,
name='cross_entropy')
tf.add_to_collection('losses', cross_entropy_mean)
correct_prediction = tf.cast(
tf.equal(tf.argmax(cosface_logits, 1),
tf.cast(label_batch, tf.int64)), tf.float32)
accuracy = tf.reduce_mean(correct_prediction)
# for weights in slim.get_variables_by_name('embedding_weights'):
# weight_regularization = tf.contrib.layers.l2_regularizer(args.weight_decay)(weights)
# tf.add_to_collection(tf.GraphKeys.REGULARIZATION_LOSSES, weight_regularization)
regularization_losses = tf.get_collection(
tf.GraphKeys.REGULARIZATION_LOSSES)
if args.weight_decay == 0:
total_loss = tf.add_n([cross_entropy_mean], name='total_loss')
else:
total_loss = tf.add_n([cross_entropy_mean] + regularization_losses,
name='total_loss')
#define two saver in case under 'finetuning on different dataset' situation
saver_save = tf.train.Saver(tf.trainable_variables(), max_to_keep=1)
train_op = facenet.train(total_loss, global_step, args.optimizer,
learning_rate, args.moving_average_decay,
tf.global_variables(), args.log_histograms)
summary_op = tf.summary.merge_all()
# Start running operations on the Graph.
gpu_options = tf.GPUOptions(
per_process_gpu_memory_fraction=args.gpu_memory_fraction)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options,
log_device_placement=False))
sess.run(tf.global_variables_initializer())
sess.run(tf.local_variables_initializer())
summary_writer = tf.summary.FileWriter(log_dir, sess.graph)
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
with sess.as_default():
if pretrained_model:
print('Restoring pretrained model: %s' % pretrained_model)
facenet.load_model(pretrained_model)
print('Running cosface training')
best_accuracy = 0.0
for epoch in range(1, args.max_nrof_epochs + 1):
step = sess.run(global_step, feed_dict=None)
cont = train(
args, sess, epoch, image_list, label_list,
index_dequeue_op, enqueue_op, image_paths_placeholder,
labels_placeholder, learning_rate_placeholder,
phase_train_placeholder, batch_size_placeholder,
global_step, total_loss, accuracy, train_op, summary_op,
summary_writer, regularization_losses,
args.learning_rate_schedule_file, args.random_rotate,
args.random_crop, args.random_flip,
args.use_fixed_image_standardization, control_placeholder)
if not cont:
break
print('validation running...')
if args.lfw_dir:
best_accuracy = evaluate(
sess, enqueue_op, image_paths_placeholder,
labels_placeholder, phase_train_placeholder,
batch_size_placeholder, control_placeholder,
embeddings, label_batch, lfw_paths, actual_issame,
args.lfw_batch_size, args.lfw_nrof_folds, log_dir,
step, summary_writer, best_accuracy, saver_save,
model_dir, subdir, args.lfw_subtract_mean,
args.lfw_use_flipped_images,
args.use_fixed_image_standardization,
args.lfw_distance_metric)
return model_dir
def main(args):
with tf.Graph().as_default():
with tf.Session() as sess:
print("args", args)
print("args.lfw_dir", args.lfw_dir)
# Read the file containing the pairs used for testing
#pairs = lfw.read_pairs(os.path.expanduser(args.lfw_pairs))
# Get the paths for the corresponding images
#paths, actual_issame = lfw.get_paths(os.path.expanduser(args.lfw_dir), pairs)
#print("paths[0:4]", len(paths), paths[0:4])
#print("actual_issame[0:4]", len(actual_issame), actual_issame[0:4])
# init paths:
## /home/kangeunsu/progressive_growing_of_gans/celeba-hq-1024x1024_test_extract/
#args.lfw_dir = "/home/kangeunsu/progressive_growing_of_gans/celeba-hq-1024x1024_test_extract"
# files like img00009982.png
files = sorted(os.listdir(args.lfw_dir + "/"))
frame_files = fnmatch.filter(files, '*.png')
full_paths = [args.lfw_dir + "/" + file for file in frame_files]
paths = full_paths #[0:12000] # full?!
actual_issame = [True] * int(len(paths) / 2) #6000 #idk
print("paths[0:4]", len(paths), paths[0:4])
print("actual_issame[0:4]", len(actual_issame), actual_issame[0:4])
# actual_issame is True and False
image_paths_placeholder = tf.placeholder(tf.string,
shape=(None, 1),
name='image_paths')
labels_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='labels')
batch_size_placeholder = tf.placeholder(tf.int32,
name='batch_size')
control_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='control')
phase_train_placeholder = tf.placeholder(tf.bool,
name='phase_train')
nrof_preprocess_threads = 4
image_size = (args.image_size, args.image_size)
print("image_size", image_size, "from args.image_size=",
args.image_size)
eval_input_queue = data_flow_ops.FIFOQueue(
capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1, ), (1, ), (1, )],
shared_name=None,
name=None)
eval_enqueue_op = eval_input_queue.enqueue_many(
[
image_paths_placeholder, labels_placeholder,
control_placeholder
],
name='eval_enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(
eval_input_queue, image_size, nrof_preprocess_threads,
batch_size_placeholder)
# Load the model
input_map = {
'image_batch': image_batch,
'label_batch': label_batch,
'phase_train': phase_train_placeholder
}
facenet.load_model(args.model, input_map=input_map)
# Get output tensor
embeddings = tf.get_default_graph().get_tensor_by_name(
"embeddings:0")
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
evaluate(sess, eval_enqueue_op, image_paths_placeholder,
labels_placeholder, phase_train_placeholder,
batch_size_placeholder, control_placeholder, embeddings,
label_batch, paths, actual_issame, args.lfw_batch_size,
args.lfw_nrof_folds, args.distance_metric,
args.subtract_mean, args.use_flipped_images,
args.use_fixed_image_standardization)
def get_embeddings(image_paths, model, batch_size):
embeddings = None
with tf.Graph().as_default():
with tf.Session() as sess:
paths = image_paths
actual_issame = None
image_paths_placeholder = tf.placeholder(tf.string,
shape=(None, 1),
name='image_paths')
labels_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='labels')
batch_size_placeholder = tf.placeholder(tf.int32,
name='batch_size')
control_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='control')
phase_train_placeholder = tf.placeholder(tf.bool,
name='phase_train')
nrof_preprocess_threads = 4
image_size = (100, 100)
eval_input_queue = data_flow_ops.FIFOQueue(
capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1, ), (1, ), (1, )],
shared_name=None,
name=None)
eval_enqueue_op = eval_input_queue.enqueue_many(
[
image_paths_placeholder, labels_placeholder,
control_placeholder
],
name='eval_enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(
eval_input_queue, image_size, nrof_preprocess_threads,
batch_size_placeholder)
# Load the model
input_map = {
'image_batch': image_batch,
'label_batch': label_batch,
'phase_train': phase_train_placeholder
}
facenet.load_model(model, input_map=input_map)
# Get output tensor
embeddings = tf.get_default_graph().get_tensor_by_name(
"embeddings:0")
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
#for the first round save the input map, so that we can visualize images of positives
embeddings = evaluate(sess, eval_enqueue_op,
image_paths_placeholder, labels_placeholder,
phase_train_placeholder,
batch_size_placeholder, control_placeholder,
embeddings, label_batch, paths, batch_size)
return embeddings
def main(args):
print('args used to train the model:')
print(args)
# =========================================================================== #
# 数据预处理,路径问题等
# =========================================================================== #
network = importlib.import_module(args.model_def)
image_size = (args.image_size, args.image_size)
subdir = datetime.strftime(datetime.now(),
'%Y%m%d-%H%M%S') # '20180614-100359'
subdir += args.dir_postfix
log_dir = os.path.join(os.path.expanduser(args.logs_base_dir),
subdir) # 本地保存log的路径
log_dir += args.dir_postfix
val_log_dir = log_dir # 本地保存validation result的路径
if not os.path.isdir(
log_dir): # Create the log directory if it doesn't exist
os.makedirs(log_dir)
model_dir = os.path.join(os.path.expanduser(args.models_base_dir),
subdir) # 本地保存model的路径,最后一层目录用上面生成的日期命名
if not os.path.isdir(
model_dir): # Create the model directory if it doesn't exist
os.makedirs(model_dir)
stat_file_name = os.path.join(log_dir, 'stat.h5')
# Write arguments to a text file, 把训练时用的所有参数保存到txt中
facenet.write_arguments_to_file(args, os.path.join(log_dir,
'arguments.txt'))
# Store some git revision info in a text file in the log directory
src_path, _ = os.path.split(os.path.realpath(__file__))
facenet.store_revision_info(src_path, log_dir, ' '.join(sys.argv))
np.random.seed(seed=args.seed)
random.seed(args.seed)
# dataset中是多个ImageClass对象,每个对象中有类名(name)和该类所有图片的路径(image_paths)
dataset = facenet.get_dataset(args.data_dir)
# filter_name default='', 这个训练的时候没用到, 先不管了
if args.filter_filename:
dataset = filter_dataset(dataset,
os.path.expanduser(args.filter_filename),
args.filter_percentile,
args.filter_min_nrof_images_per_class)
# validation_set_split_ratio default=0.0, 如果需要验证操作的话设置这个
# 划分验证集有两种方式:按类划分和按图片划分(默认)
if args.validation_set_split_ratio > 0.0:
train_set, val_set = facenet.split_dataset(
dataset, args.validation_set_split_ratio,
args.min_nrof_val_images_per_class, 'SPLIT_IMAGES')
else:
train_set, val_set = dataset, []
nrof_classes = len(train_set)
print('Model directory: %s' % model_dir)
print('Log directory: %s' % log_dir)
pretrained_model = None
if args.pretrained_model:
pretrained_model = os.path.expanduser(args.pretrained_model)
print('Pre-trained model: %s' % pretrained_model)
if args.lfw_dir:
print('LFW directory: %s' % args.lfw_dir)
# Read the file containing the pairs used for testing
# pairs[0] = ['Abel_Pacheco', '1', '4']
# args.lfw_pairs: data/pairs.txt, 这个文件中每一行或有三个字段或有四个字段
# 如果有三个字段表示从一个人的所有图片中选择两个(正样本, actual_issame=True)
# 如果有四个字段表示从两个人的所有图片中各选一个(负样本, actual_issame=False)
pairs = lfw.read_pairs(os.path.expanduser(args.lfw_pairs))
# Get the paths for the corresponding images
# lfw_paths里的每个值都是一个元组(path0, path1), actual_issame里对应的是这个元组是正样本还是副样本
lfw_paths, actual_issame = lfw.get_paths(
os.path.expanduser(args.lfw_dir), pairs)
# =========================================================================== #
# 下面开始处理TensorFlow相关的东西
# =========================================================================== #
with tf.Graph().as_default():
# =========================================================================== #
# Input pipeline
# =========================================================================== #
tf.set_random_seed(args.seed)
global_step = tf.Variable(0, trainable=False)
# Get a list of image paths and their labels
# 这两个list是训练集中的所有数据
image_list, label_list = facenet.get_image_paths_and_labels(train_set)
assert len(image_list) > 0, 'The training set should not be empty'
val_image_list, val_label_list = facenet.get_image_paths_and_labels(
val_set)
# Create a queue that produces indices into the image_list and label_list
labels = ops.convert_to_tensor(label_list, dtype=tf.int32)
range_size = array_ops.shape(labels)[0]
# num_epochs=None时,循环生成0到range_size-1的数
index_queue = tf.train.range_input_producer(range_size,
num_epochs=None,
shuffle=True,
seed=None,
capacity=32)
# 这个op是用来生成索引值的,每次从队列里取出一个epoch中所有图片数量的下标
index_dequeue_op = index_queue.dequeue_many(
args.batch_size * args.epoch_size, 'index_dequeue')
learning_rate_placeholder = tf.placeholder(tf.float32,
name='learning_rate')
batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size')
phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train')
image_paths_placeholder = tf.placeholder(tf.string,
shape=(None, 1),
name='image_paths')
labels_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='labels')
control_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='control')
nrof_preprocess_threads = 4
input_queue = data_flow_ops.FIFOQueue(
capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1, ), (1, ), (1, )],
shared_name=None,
name=None)
enqueue_op = input_queue.enqueue_many(
[image_paths_placeholder, labels_placeholder, control_placeholder],
name='enqueue_op')
# create input pipeline, 将上面image path中的文件取出来
image_batch, label_batch = facenet.create_input_pipeline(
input_queue, image_size, nrof_preprocess_threads,
batch_size_placeholder)
image_batch = tf.identity(image_batch, 'image_batch')
image_batch = tf.identity(image_batch, 'input')
label_batch = tf.identity(label_batch, 'label_batch')
print('Number of classes in training set: %d' % nrof_classes)
print('Number of examples in training set: %d' % len(image_list))
print('Number of classes in validation set: %d' % len(val_set))
print('Number of examples in validation set: %d' % len(val_image_list))
print('Building training graph')
### Build the inference graph
# prelogits是128维的向量
prelogits, _ = network.inference(
image_batch,
args.keep_probability,
phase_train=phase_train_placeholder,
bottleneck_layer_size=args.embedding_size,
weight_decay=args.weight_decay)
# TODO: Inject some code here for fine tuning
# logits的shape: [None, len(train_set)]
logits = slim.fully_connected(
prelogits,
len(train_set),
activation_fn=None,
weights_initializer=slim.initializers.xavier_initializer(),
weights_regularizer=slim.l2_regularizer(args.weight_decay),
scope='Logits',
reuse=False)
# 这步操作之后在测试时就可以拿到这个值了
logits = tf.identity(logits, name='logits')
# axis=1, 是把每行进行normalize, 每行是一个样本, 128维(或512维)的向量
embeddings = tf.nn.l2_normalize(prelogits, 1, 1e-10, name='embeddings')
# Norm for the prelogits
eps = 1e-4
prelogits_norm = tf.reduce_mean(
tf.norm(tf.abs(prelogits) + eps, ord=args.prelogits_norm_p,
axis=1))
# 把prelogits进行norm操作得到的是一个数,这个数乘个weight作为loss的一部分,--prelogits_norm_loss_factor 5e-4
tf.add_to_collection(tf.GraphKeys.REGULARIZATION_LOSSES,
prelogits_norm * args.prelogits_norm_loss_factor)
# Add center loss, 这个center loss作为正则化loss的一部分, 出自一篇人脸识别的论文, 先不管
# 这个loss默认权重是0, 文档中给出的训练脚本并没有用这个loss
prelogits_center_loss, _ = facenet.center_loss(prelogits, label_batch,
args.center_loss_alfa,
nrof_classes)
tf.add_to_collection(tf.GraphKeys.REGULARIZATION_LOSSES,
prelogits_center_loss * args.center_loss_factor)
learning_rate = tf.train.exponential_decay(
learning_rate_placeholder,
global_step,
args.learning_rate_decay_epochs * args.epoch_size,
args.learning_rate_decay_factor,
staircase=True)
tf.summary.scalar('learning_rate', learning_rate)
# Calculate the average cross entropy loss across the batch
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(
labels=label_batch,
logits=logits,
name='cross_entropy_per_example')
cross_entropy_mean = tf.reduce_mean(cross_entropy,
name='cross_entropy')
tf.add_to_collection('losses', cross_entropy_mean)
correct_prediction = tf.cast(
tf.equal(tf.argmax(logits, 1), tf.cast(label_batch, tf.int64)),
tf.float32)
accuracy = tf.reduce_mean(correct_prediction)
# Calculate the total losses
regularization_losses = tf.get_collection(
tf.GraphKeys.REGULARIZATION_LOSSES)
# 这个total_loss是ce和regular loss的和,应该是已经把所有loss的值都加起来了
total_loss = tf.add_n([cross_entropy_mean] + regularization_losses,
name='total_loss')
# Build a Graph that trains the model with one batch of examples and updates the model parameters
train_op = facenet.train(total_loss, global_step, args.optimizer,
learning_rate, args.moving_average_decay,
tf.global_variables(), args.log_histograms)
# Create a saver
saver = tf.train.Saver(tf.trainable_variables(), max_to_keep=20)
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.summary.merge_all()
### Start running operations on the Graph.
gpu_options = tf.GPUOptions(
per_process_gpu_memory_fraction=args.gpu_memory_fraction)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options,
log_device_placement=False))
sess.run(tf.global_variables_initializer())
sess.run(tf.local_variables_initializer())
summary_writer = tf.summary.FileWriter(log_dir, sess.graph)
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
# =========================================================================== #
# Start training
# =========================================================================== #
with sess.as_default():
if pretrained_model:
print('Restoring pretrained model: %s' % pretrained_model)
# saver.restore(sess, pretrained_model)
if args.checkpoint_exclude_scopes:
exclusions = [args.checkpoint_exclude_scopes.split(',')]
else:
exclusions = []
# load from pre-trained model
tf.contrib.framework.assign_from_checkpoint_fn(
pretrained_model, [
var for var in tf.trainable_variables()
if all(not var.op.name.startswith(exclusion)
for exclusion in exclusions)
],
ignore_missing_vars=args.ignore_missing_vars)(sess)
# Training and validation loop
print('Running training')
# max_nrof_epochs: number of epochs to run, default=500
# epoch_size: number of batches per epoch,default=1000
nrof_steps = args.max_nrof_epochs * args.epoch_size # total batch number
# 在验证集上验证的次数
nrof_val_samples = int(
math.ceil(args.max_nrof_epochs / args.validate_every_n_epochs)
) # Validate every validate_every_n_epochs as well as in the last epoch
stat = {
# 每一个batch记录一次
'loss':
np.zeros((nrof_steps, ), np.float32),
'center_loss':
np.zeros((nrof_steps, ), np.float32),
'reg_loss':
np.zeros((nrof_steps, ), np.float32),
'xent_loss':
np.zeros((nrof_steps, ), np.float32),
'prelogits_norm':
np.zeros((nrof_steps, ), np.float32),
'accuracy':
np.zeros((nrof_steps, ), np.float32),
# 每验证一次记录一次
'val_loss':
np.zeros((nrof_val_samples, ), np.float32),
'val_xent_loss':
np.zeros((nrof_val_samples, ), np.float32),
'val_accuracy':
np.zeros((nrof_val_samples, ), np.float32),
# 每个epoch记录一次
'lfw_accuracy':
np.zeros((args.max_nrof_epochs, ), np.float32),
'lfw_valrate':
np.zeros((args.max_nrof_epochs, ), np.float32),
'learning_rate':
np.zeros((args.max_nrof_epochs, ), np.float32),
'time_train':
np.zeros((args.max_nrof_epochs, ), np.float32),
'time_validate':
np.zeros((args.max_nrof_epochs, ), np.float32),
'time_evaluate':
np.zeros((args.max_nrof_epochs, ), np.float32),
'prelogits_hist':
np.zeros((args.max_nrof_epochs, 1000), np.float32),
}
for epoch in range(1, args.max_nrof_epochs + 1):
step = sess.run(global_step, feed_dict=None) # step = 0
### Train for one epoch
t = time.time()
# =========================================================================== #
# 一次训练的过程,学习率是在这里设置的,如果参数中的学习率小于0,就用调度文件中的学习率
# =========================================================================== #
cont = train(
args, sess, epoch, image_list, label_list,
index_dequeue_op, enqueue_op, image_paths_placeholder,
labels_placeholder, learning_rate_placeholder,
phase_train_placeholder, batch_size_placeholder,
control_placeholder, global_step, total_loss, train_op,
summary_op, summary_writer, regularization_losses,
args.learning_rate_schedule_file, stat, cross_entropy_mean,
accuracy, learning_rate, prelogits, prelogits_center_loss,
args.random_rotate, args.random_crop, args.random_flip,
prelogits_norm, args.prelogits_hist_max,
args.use_fixed_image_standardization)
stat['time_train'][epoch - 1] = time.time() - t
if not cont:
# 每次结束一轮训练,train函数返回True
break
### validate on val_set
t = time.time()
if len(val_image_list) > 0 and \
((
epoch - 1) % args.validate_every_n_epochs == args.validate_every_n_epochs - 1 or epoch == args.max_nrof_epochs):
# =========================================================================== #
# 如果验证集不为空 并且 到了设置的每n个epoch验证一次或者到了最后一个epoch, 则进行一次验证
# =========================================================================== #
validate(args, sess, epoch, val_image_list, val_label_list,
enqueue_op, image_paths_placeholder,
labels_placeholder, control_placeholder,
phase_train_placeholder, batch_size_placeholder,
stat, total_loss, regularization_losses,
cross_entropy_mean, accuracy,
args.validate_every_n_epochs,
args.use_fixed_image_standardization, val_log_dir)
stat['time_validate'][epoch - 1] = time.time() - t
# Save variables and the metagraph if it doesn't exist
save_variables_and_metagraph(sess, saver, summary_writer,
model_dir, subdir, epoch)
### Evaluate on LFW
t = time.time()
if args.lfw_dir:
evaluate(sess, enqueue_op, image_paths_placeholder,
labels_placeholder, phase_train_placeholder,
batch_size_placeholder, control_placeholder,
args.lfw_nrof_folds, log_dir, step,
summary_writer, stat, epoch, embeddings,
label_batch, lfw_paths, actual_issame,
args.lfw_batch_size, args.lfw_distance_metric,
args.lfw_subtract_mean,
args.lfw_use_flipped_images,
args.use_fixed_image_standardization)
stat['time_evaluate'][epoch - 1] = time.time() - t
print('Saving statistics')
with h5py.File(stat_file_name, 'w') as f:
for key, value in stat.items():
f.create_dataset(key, data=value)
return model_dir
def main(args):
network = importlib.import_module(args.model_def)
image_size = (args.image_size, args.image_size)
subdir = datetime.strftime(datetime.now(), '%Y%m%d-%H%M%S')
log_dir = os.path.join(os.path.expanduser(args.logs_base_dir), subdir)
if not os.path.isdir(log_dir): # Create the log directory if it doesn't exist
os.makedirs(log_dir)
model_dir = os.path.join(os.path.expanduser(args.models_base_dir), subdir)
if not os.path.isdir(model_dir): # Create the model directory if it doesn't exist
os.makedirs(model_dir)
stat_file_name = os.path.join(log_dir, 'stat.h5')
# Write arguments to a text file
facenet.write_arguments_to_file(args, os.path.join(log_dir, 'arguments.txt'))
# Store some git revision info in a text file in the log directory
src_path,_ = os.path.split(os.path.realpath(__file__))
facenet.store_revision_info(src_path, log_dir, ' '.join(sys.argv))
np.random.seed(seed=args.seed)
random.seed(args.seed)
dataset = facenet.get_dataset(args.data_dir)
if args.filter_filename:
dataset = filter_dataset(dataset, os.path.expanduser(args.filter_filename),
args.filter_percentile, args.filter_min_nrof_images_per_class)
if args.validation_set_split_ratio>0.0:
train_set, val_set = facenet.split_dataset(dataset, args.validation_set_split_ratio, args.min_nrof_val_images_per_class, 'SPLIT_IMAGES')
else:
train_set, val_set = dataset, []
nrof_classes = len(train_set)
print('Model directory: %s' % model_dir)
print('Log directory: %s' % log_dir)
pretrained_model = None
if args.pretrained_model:
pretrained_model = os.path.expanduser(args.pretrained_model)
print('Pre-trained model: %s' % pretrained_model)
if args.lfw_dir:
print('LFW directory: %s' % args.lfw_dir)
# Read the file containing the pairs used for testing
pairs = lfw.read_pairs(os.path.expanduser(args.lfw_pairs))
# Get the paths for the corresponding images
lfw_paths, actual_issame = lfw.get_paths(os.path.expanduser(args.lfw_dir), pairs)
with tf.Graph().as_default():
tf.set_random_seed(args.seed)
global_step = tf.Variable(0, trainable=False)
# Get a list of image paths and their labels
image_list, label_list = facenet.get_image_paths_and_labels(train_set)
assert len(image_list)>0, 'The training set should not be empty'
val_image_list, val_label_list = facenet.get_image_paths_and_labels(val_set)
# Create a queue that produces indices into the image_list and label_list
labels = ops.convert_to_tensor(label_list, dtype=tf.int32)
range_size = array_ops.shape(labels)[0]
index_queue = tf.train.range_input_producer(range_size, num_epochs=None,
shuffle=True, seed=None, capacity=32)
index_dequeue_op = index_queue.dequeue_many(args.batch_size*args.epoch_size, 'index_dequeue')
learning_rate_placeholder = tf.placeholder(tf.float32, name='learning_rate')
batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size')
phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train')
image_paths_placeholder = tf.placeholder(tf.string, shape=(None,1), name='image_paths')
labels_placeholder = tf.placeholder(tf.int32, shape=(None,1), name='labels')
control_placeholder = tf.placeholder(tf.int32, shape=(None,1), name='control')
nrof_preprocess_threads = 4
input_queue = data_flow_ops.FIFOQueue(capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1,), (1,), (1,)],
shared_name=None, name=None)
enqueue_op = input_queue.enqueue_many([image_paths_placeholder, labels_placeholder, control_placeholder], name='enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(input_queue, image_size, nrof_preprocess_threads, batch_size_placeholder)
image_batch = tf.identity(image_batch, 'image_batch')
image_batch = tf.identity(image_batch, 'input')
label_batch = tf.identity(label_batch, 'label_batch')
print('Number of classes in training set: %d' % nrof_classes)
print('Number of examples in training set: %d' % len(image_list))
print('Number of classes in validation set: %d' % len(val_set))
print('Number of examples in validation set: %d' % len(val_image_list))
print('Building training graph')
# Build the inference graph
prelogits, _ = network.inference(image_batch, args.keep_probability,
phase_train=phase_train_placeholder, bottleneck_layer_size=args.embedding_size,
weight_decay=args.weight_decay)
logits = slim.fully_connected(prelogits, len(train_set), activation_fn=None,
weights_initializer=slim.initializers.xavier_initializer(),
weights_regularizer=slim.l2_regularizer(args.weight_decay),
scope='Logits', reuse=False)
embeddings = tf.nn.l2_normalize(prelogits, 1, 1e-10, name='embeddings')
# Norm for the prelogits
eps = 1e-4
prelogits_norm = tf.reduce_mean(tf.norm(tf.abs(prelogits)+eps, ord=args.prelogits_norm_p, axis=1))
tf.add_to_collection(tf.GraphKeys.REGULARIZATION_LOSSES, prelogits_norm * args.prelogits_norm_loss_factor)
# Add center loss
prelogits_center_loss, _ = facenet.center_loss(prelogits, label_batch, args.center_loss_alfa, nrof_classes)
tf.add_to_collection(tf.GraphKeys.REGULARIZATION_LOSSES, prelogits_center_loss * args.center_loss_factor)
learning_rate = tf.train.exponential_decay(learning_rate_placeholder, global_step,
args.learning_rate_decay_epochs*args.epoch_size, args.learning_rate_decay_factor, staircase=True)
tf.summary.scalar('learning_rate', learning_rate)
# Calculate the average cross entropy loss across the batch
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(
labels=label_batch, logits=logits, name='cross_entropy_per_example')
cross_entropy_mean = tf.reduce_mean(cross_entropy, name='cross_entropy')
tf.add_to_collection('losses', cross_entropy_mean)
correct_prediction = tf.cast(tf.equal(tf.argmax(logits, 1), tf.cast(label_batch, tf.int64)), tf.float32)
accuracy = tf.reduce_mean(correct_prediction)
# Calculate the total losses
regularization_losses = tf.get_collection(tf.GraphKeys.REGULARIZATION_LOSSES)
total_loss = tf.add_n([cross_entropy_mean] + regularization_losses, name='total_loss')
# Build a Graph that trains the model with one batch of examples and updates the model parameters
train_op = facenet.train(total_loss, global_step, args.optimizer,
learning_rate, args.moving_average_decay, tf.global_variables(), args.log_histograms)
# Create a saver
saver = tf.train.Saver(tf.trainable_variables(), max_to_keep=3)
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.summary.merge_all()
# Start running operations on the Graph.
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=args.gpu_memory_fraction)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options, log_device_placement=False))
sess.run(tf.global_variables_initializer())
sess.run(tf.local_variables_initializer())
summary_writer = tf.summary.FileWriter(log_dir, sess.graph)
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
with sess.as_default():
if pretrained_model:
print('Restoring pretrained model: %s' % pretrained_model)
saver.restore(sess, pretrained_model)
# Training and validation loop
print('Running training')
nrof_steps = args.max_nrof_epochs*args.epoch_size
nrof_val_samples = int(math.ceil(args.max_nrof_epochs / args.validate_every_n_epochs)) # Validate every validate_every_n_epochs as well as in the last epoch
stat = {
'loss': np.zeros((nrof_steps,), np.float32),
'center_loss': np.zeros((nrof_steps,), np.float32),
'reg_loss': np.zeros((nrof_steps,), np.float32),
'xent_loss': np.zeros((nrof_steps,), np.float32),
'prelogits_norm': np.zeros((nrof_steps,), np.float32),
'accuracy': np.zeros((nrof_steps,), np.float32),
'val_loss': np.zeros((nrof_val_samples,), np.float32),
'val_xent_loss': np.zeros((nrof_val_samples,), np.float32),
'val_accuracy': np.zeros((nrof_val_samples,), np.float32),
'lfw_accuracy': np.zeros((args.max_nrof_epochs,), np.float32),
'lfw_valrate': np.zeros((args.max_nrof_epochs,), np.float32),
'learning_rate': np.zeros((args.max_nrof_epochs,), np.float32),
'time_train': np.zeros((args.max_nrof_epochs,), np.float32),
'time_validate': np.zeros((args.max_nrof_epochs,), np.float32),
'time_evaluate': np.zeros((args.max_nrof_epochs,), np.float32),
'prelogits_hist': np.zeros((args.max_nrof_epochs, 1000), np.float32),
}
for epoch in range(1,args.max_nrof_epochs+1):
step = sess.run(global_step, feed_dict=None)
# Train for one epoch
t = time.time()
cont = train(args, sess, epoch, image_list, label_list, index_dequeue_op, enqueue_op, image_paths_placeholder, labels_placeholder,
learning_rate_placeholder, phase_train_placeholder, batch_size_placeholder, control_placeholder, global_step,
total_loss, train_op, summary_op, summary_writer, regularization_losses, args.learning_rate_schedule_file,
stat, cross_entropy_mean, accuracy, learning_rate,
prelogits, prelogits_center_loss, args.random_rotate, args.random_crop, args.random_flip, prelogits_norm, args.prelogits_hist_max, args.use_fixed_image_standardization)
stat['time_train'][epoch-1] = time.time() - t
if not cont:
break
t = time.time()
if len(val_image_list)>0 and ((epoch-1) % args.validate_every_n_epochs == args.validate_every_n_epochs-1 or epoch==args.max_nrof_epochs):
validate(args, sess, epoch, val_image_list, val_label_list, enqueue_op, image_paths_placeholder, labels_placeholder, control_placeholder,
phase_train_placeholder, batch_size_placeholder,
stat, total_loss, regularization_losses, cross_entropy_mean, accuracy, args.validate_every_n_epochs, args.use_fixed_image_standardization)
stat['time_validate'][epoch-1] = time.time() - t
# Save variables and the metagraph if it doesn't exist already
save_variables_and_metagraph(sess, saver, summary_writer, model_dir, subdir, epoch)
# Evaluate on LFW
t = time.time()
if args.lfw_dir:
evaluate(sess, enqueue_op, image_paths_placeholder, labels_placeholder, phase_train_placeholder, batch_size_placeholder, control_placeholder,
embeddings, label_batch, lfw_paths, actual_issame, args.lfw_batch_size, args.lfw_nrof_folds, log_dir, step, summary_writer, stat, epoch,
args.lfw_distance_metric, args.lfw_subtract_mean, args.lfw_use_flipped_images, args.use_fixed_image_standardization)
stat['time_evaluate'][epoch-1] = time.time() - t
print('Saving statistics')
with h5py.File(stat_file_name, 'w') as f:
for key, value in stat.iteritems():
f.create_dataset(key, data=value)
return model_dir
def main(args):
with tf.Graph().as_default():
with tf.Session() as sess:
image_dir1 = os.path.expanduser(args.first_dir)
image_dir2 = os.path.expanduser(args.second_dir)
image_paths_placeholder = tf.placeholder(tf.string,
shape=(None, 1),
name='image_paths')
labels_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='labels')
batch_size_placeholder = tf.placeholder(tf.int32,
name='batch_size')
control_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='control')
phase_train_placeholder = tf.placeholder(tf.bool,
name='phase_train')
nrof_preprocess_threads = 4
image_size = (args.image_size, args.image_size)
eval_input_queue = data_flow_ops.FIFOQueue(
capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1, ), (1, ), (1, )],
shared_name=None,
name=None)
eval_enqueue_op = eval_input_queue.enqueue_many(
[
image_paths_placeholder, labels_placeholder,
control_placeholder
],
name='eval_enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(
eval_input_queue, image_size, nrof_preprocess_threads,
batch_size_placeholder)
# Load the model
input_map = {
'image_batch': image_batch,
'label_batch': label_batch,
'phase_train': phase_train_placeholder
}
print('=== model: ')
print(args.model)
print('===input_map: ')
print(input_map)
facenet.load_model(args.model, input_map=input_map)
images_placeholder = tf.get_default_graph().get_tensor_by_name(
"input:0")
# Get output tensor
embeddings = tf.get_default_graph().get_tensor_by_name(
"embeddings:0")
embedding_size = embeddings.get_shape()[1]
print('facenet embedding模型建立完毕')
for num in range(0, args.image_num):
scaled_reshape = []
image_name1 = "%s/%d.jpg" % (image_dir1, num)
image_name2 = "%s/%d.jpg" % (image_dir2, num)
image1 = scipy.misc.imread(image_name1, mode='RGB')
image1 = cv2.resize(image1, (args.image_size, args.image_size),
interpolation=cv2.INTER_CUBIC)
image1 = facenet.prewhiten(image1)
scaled_reshape.append(
image1.reshape(-1, args.image_size, args.image_size, 3))
emb_array1 = np.zeros((1, embedding_size))
emb_array1[0, :] = sess.run(embeddings,
feed_dict={
images_placeholder:
scaled_reshape[0],
phase_train_placeholder:
False
})[0]
image2 = scipy.misc.imread(image_name2, mode='RGB')
image2 = cv2.resize(image2, (args.image_size, args.image_size),
interpolation=cv2.INTER_CUBIC)
image2 = facenet.prewhiten(image2)
scaled_reshape.append(
image2.reshape(-1, args.image_size, args.image_size, 3))
emb_array2 = np.zeros((1, embedding_size))
emb_array2[0, :] = sess.run(embeddings,
feed_dict={
images_placeholder:
scaled_reshape[1],
phase_train_placeholder:
False
})[0]
dist = np.sqrt(np.sum(np.square(emb_array1[0] -
emb_array2[0])))
print("第%d组照片特征向量的欧氏距离：%f " % (num, dist))
if dist < 1.028:
print(" 第%d组照片是同一个人 " % num)
else:
print(" 第%d组照片不是同一个人 " % num)
def generate_data(feature_size, max_length, batch_size, MAT_folder, img_folder):
# load mat files
Mat_paths = os.listdir(MAT_folder)
choose_idx = np.random.randint(len(Mat_paths), size=batch_size)
Mat_files = []
seq_names = []
for n in range(batch_size):
seq_name = Mat_paths[choose_idx[n]][0:21]+'_sync'
temp_path = data_folder+'/'+seq_name+'/gt_2.mat'
temp_mat_file = loadmat(temp_path)
Mat_files.append(temp_mat_file)
seq_names.append(seq_name)
X = np.zeros((batch_size,feature_size,max_length,3))
Y = np.zeros((batch_size,2))
all_paths = []
# positive
for n in range(int(batch_size/2)):
seq_name = seq_names[n]
fr_num = Mat_files[n]['gtInfo'][0][0][0].shape[0]
id_num = Mat_files[n]['gtInfo'][0][0][0].shape[1]
Y[n,0] = 1
X_3d = Mat_files[n]['gtInfo'][0][0][4]
Y_3d = Mat_files[n]['gtInfo'][0][0][5]
W_3d = Mat_files[n]['gtInfo'][0][0][6]
H_3d = Mat_files[n]['gtInfo'][0][0][7]
try_time = 0
if try_time>=10:
continue
while 1:
if try_time>=10:
all_paths.append([])
#print('err')
break
obj_id = np.random.randint(id_num, size=1)[0]
part_W_mat = Mat_files[n]['gtInfo'][0][0][3][:,obj_id]
non_zero_idx = np.where(part_W_mat>0)[0]
if np.max(non_zero_idx)-np.min(non_zero_idx)+1!=len(non_zero_idx) or len(non_zero_idx)<=1:
try_time = try_time+1
continue
st_fr = np.min(non_zero_idx)#+np.random.randint(len(non_zero_idx)-1, size=1)[0]
end_fr = np.max(non_zero_idx)
abs_fr_t1 = int(st_fr+np.random.randint(len(non_zero_idx)-1, size=1)[0])
abs_end_fr = min(abs_fr_t1+max_length-1,end_fr)
abs_fr_t4 = int(abs_end_fr-np.random.randint(abs_end_fr-abs_fr_t1, size=1)[0])
abs_fr_t2 = int(abs_fr_t1+np.random.randint(abs_fr_t4-abs_fr_t1, size=1)[0])
abs_fr_t3 = int(abs_fr_t4-np.random.randint(abs_fr_t4-abs_fr_t2, size=1)[0])
t1 = 0
t2 = abs_fr_t2-abs_fr_t1
t3 = abs_fr_t3-abs_fr_t1
t4 = abs_fr_t4-abs_fr_t1
# mask
X[n,:,t1:t2+1,1] = 1
X[n,:,t3:t4+1,2] = 1
# X
X[n,0,t1:t2+1,0] = X_3d[abs_fr_t1:abs_fr_t2+1,obj_id]/loc_scales[0]+noise_scales[0]*np.random.normal(0,1,t2-t1+1)
X[n,0,t3:t4+1,0] = X_3d[abs_fr_t3:abs_fr_t4+1,obj_id]/loc_scales[0]+noise_scales[0]*np.random.normal(0,1,t4-t3+1)
# Y
X[n,1,t1:t2+1,0] = Y_3d[abs_fr_t1:abs_fr_t2+1,obj_id]/loc_scales[1]+noise_scales[1]*np.random.normal(0,1,t2-t1+1)
X[n,1,t3:t4+1,0] = Y_3d[abs_fr_t3:abs_fr_t4+1,obj_id]/loc_scales[1]+noise_scales[1]*np.random.normal(0,1,t4-t3+1)
# W
X[n,2,t1:t2+1,0] = W_3d[abs_fr_t1:abs_fr_t2+1,obj_id]/loc_scales[2]+noise_scales[2]*np.random.normal(0,1,t2-t1+1)
X[n,2,t3:t4+1,0] = W_3d[abs_fr_t3:abs_fr_t4+1,obj_id]/loc_scales[2]+noise_scales[2]*np.random.normal(0,1,t4-t3+1)
# H
X[n,3,t1:t2+1,0] = H_3d[abs_fr_t1:abs_fr_t2+1,obj_id]/loc_scales[3]+noise_scales[3]*np.random.normal(0,1,t2-t1+1)
X[n,3,t3:t4+1,0] = H_3d[abs_fr_t3:abs_fr_t4+1,obj_id]/loc_scales[3]+noise_scales[3]*np.random.normal(0,1,t4-t3+1)
'''
plt.plot(X[n,0,:,0], 'ro')
plt.show()
plt.plot(X[n,1,:,0], 'ro')
plt.show()
plt.plot(X[n,2,:,0], 'ro')
plt.show()
plt.plot(X[n,3,:,0], 'ro')
plt.show()
plt.plot(X[n,0,:,1], 'ro')
plt.show()
plt.plot(X[n,0,:,2], 'ro')
plt.show()
import pdb; pdb.set_trace()
'''
temp_paths = []
for k in range(abs_fr_t1,abs_fr_t2+1):
class_name = seq_name+'_image_02_'+num_str(obj_id+1,4)
file_name = class_name+'_'+num_str(k+1,4)+'.png'
temp_path = img_folder+'/'+class_name+'/'+file_name
temp_paths.append(temp_path)
for k in range(abs_fr_t3,abs_fr_t4+1):
class_name = seq_name+'_image_02_'+num_str(obj_id+1,4)
file_name = class_name+'_'+num_str(k+1,4)+'.png'
temp_path = img_folder+'/'+class_name+'/'+file_name
temp_paths.append(temp_path)
all_paths.append(temp_paths.copy())
break
# negative
for n in range(int(batch_size/2),batch_size):
Y[n,1] = 1
seq_name = seq_names[n]
fr_num = Mat_files[n]['gtInfo'][0][0][0].shape[0]
id_num = Mat_files[n]['gtInfo'][0][0][0].shape[1]
X_3d = Mat_files[n]['gtInfo'][0][0][4]
Y_3d = Mat_files[n]['gtInfo'][0][0][5]
W_3d = Mat_files[n]['gtInfo'][0][0][6]
H_3d = Mat_files[n]['gtInfo'][0][0][7]
try_time = 0
time_interval = np.zeros((id_num,2))
for obj_id in range(id_num):
part_W_mat = Mat_files[n]['gtInfo'][0][0][3][:,obj_id]
non_zero_idx = np.where(part_W_mat>0)[0]
t_min = np.min(non_zero_idx)
t_max = np.max(non_zero_idx)
if len(non_zero_idx)!=t_max-t_min+1:
time_interval[obj_id,0] = -1
time_interval[obj_id,1] = -1
else:
time_interval[obj_id,0] = t_min
time_interval[obj_id,1] = t_max
if try_time>=10:
continue
while 1:
if try_time>=10:
all_paths.append([])
break
split_fr = 1+np.random.randint(fr_num-2, size=1)[0]
cand_pairs = []
for id1 in range(id_num):
for id2 in range(id_num):
if id1==id2:
continue
if time_interval[id1,0]==-1 or time_interval[id2,0]==-1:
continue
if time_interval[id1,0]<=split_fr and time_interval[id2,1]>split_fr:
t_above = min(split_fr,time_interval[id1,1])
t_below = max(split_fr+1,time_interval[id2,0])
t_dist = t_below-t_above
if t_dist<max_length/4:
cand_pairs.append([id1,id2,t_dist])
if len(cand_pairs)==0:
try_time = try_time+1
continue
choose_pair_idx = np.random.randint(len(cand_pairs), size=1)[0]
obj_id1 = cand_pairs[choose_pair_idx][0]
obj_id2 = cand_pairs[choose_pair_idx][1]
t_below = max(split_fr+1,time_interval[obj_id2,0])
t_above = min(split_fr,time_interval[obj_id1,1])
t_min = max(t_below-max_length+1,time_interval[obj_id1,0])
abs_fr_t1 = int(t_min+np.random.randint(t_above-t_min+1, size=1)[0])
#abs_fr_t2 = int(abs_fr_t1+np.random.randint(t_above-abs_fr_t1+1, size=1)[0])
abs_fr_t2 = int(t_above)
abs_fr_t4 = min(abs_fr_t1+max_length-1,time_interval[obj_id2,1])
abs_fr_t4 = int(abs_fr_t4-np.random.randint(abs_fr_t4-t_below+1, size=1)[0])
#abs_fr_t3 = int(abs_fr_t4-np.random.randint(abs_fr_t4-t_below+1, size=1)[0])
abs_fr_t3 = int(t_below)
'''
print(abs_fr_t1)
print(abs_fr_t2)
print(abs_fr_t3)
print(abs_fr_t4)
#import pdb; pdb.set_trace()
'''
t1 = 0
t2 = abs_fr_t2-abs_fr_t1
t3 = abs_fr_t3-abs_fr_t1
t4 = abs_fr_t4-abs_fr_t1
# mask
X[n,:,t1:t2+1,1] = 1
X[n,:,t3:t4+1,2] = 1
# X
X[n,0,t1:t2+1,0] = X_3d[abs_fr_t1:abs_fr_t2+1,obj_id1]/loc_scales[0]+noise_scales[0]*np.random.normal(0,1,t2-t1+1)
X[n,0,t3:t4+1,0] = X_3d[abs_fr_t3:abs_fr_t4+1,obj_id2]/loc_scales[0]+noise_scales[0]*np.random.normal(0,1,t4-t3+1)
# Y
X[n,1,t1:t2+1,0] = Y_3d[abs_fr_t1:abs_fr_t2+1,obj_id1]/loc_scales[1]+noise_scales[1]*np.random.normal(0,1,t2-t1+1)
X[n,1,t3:t4+1,0] = Y_3d[abs_fr_t3:abs_fr_t4+1,obj_id2]/loc_scales[1]+noise_scales[1]*np.random.normal(0,1,t4-t3+1)
# W
X[n,2,t1:t2+1,0] = W_3d[abs_fr_t1:abs_fr_t2+1,obj_id1]/loc_scales[2]+noise_scales[2]*np.random.normal(0,1,t2-t1+1)
X[n,2,t3:t4+1,0] = W_3d[abs_fr_t3:abs_fr_t4+1,obj_id2]/loc_scales[2]+noise_scales[2]*np.random.normal(0,1,t4-t3+1)
# H
X[n,3,t1:t2+1,0] = H_3d[abs_fr_t1:abs_fr_t2+1,obj_id1]/loc_scales[3]+noise_scales[3]*np.random.normal(0,1,t2-t1+1)
X[n,3,t3:t4+1,0] = H_3d[abs_fr_t3:abs_fr_t4+1,obj_id2]/loc_scales[3]+noise_scales[3]*np.random.normal(0,1,t4-t3+1)
'''
plt.plot(X[n,0,:,0], 'ro')
plt.show()
plt.plot(X[n,1,:,0], 'ro')
plt.show()
plt.plot(X[n,2,:,0], 'ro')
plt.show()
plt.plot(X[n,3,:,0], 'ro')
plt.show()
plt.plot(X[n,0,:,1], 'ro')
plt.show()
plt.plot(X[n,0,:,2], 'ro')
plt.show()
import pdb; pdb.set_trace()
'''
temp_paths = []
for k in range(abs_fr_t1,abs_fr_t2+1):
class_name = seq_name+'_image_02_'+num_str(obj_id1+1,4)
file_name = class_name+'_'+num_str(k+1,4)+'.png'
temp_path = img_folder+'/'+class_name+'/'+file_name
temp_paths.append(temp_path)
for k in range(abs_fr_t3,abs_fr_t4+1):
class_name = seq_name+'_image_02_'+num_str(obj_id2+1,4)
file_name = class_name+'_'+num_str(k+1,4)+'.png'
temp_path = img_folder+'/'+class_name+'/'+file_name
temp_paths.append(temp_path)
all_paths.append(temp_paths.copy())
break
f_image_size = 160
distance_metric = 0
with tf.Graph().as_default():
with tf.Session() as sess:
#import pdb; pdb.set_trace()
image_paths_placeholder = tf.placeholder(tf.string, shape=(None,1), name='image_paths')
labels_placeholder = tf.placeholder(tf.int32, shape=(None,1), name='labels')
batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size')
control_placeholder = tf.placeholder(tf.int32, shape=(None,1), name='control')
phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train')
nrof_preprocess_threads = 4
image_size = (f_image_size, f_image_size)
eval_input_queue = data_flow_ops.FIFOQueue(capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1,), (1,), (1,)],
shared_name=None, name=None)
eval_enqueue_op = eval_input_queue.enqueue_many([image_paths_placeholder, labels_placeholder, control_placeholder], name='eval_enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(eval_input_queue, image_size, nrof_preprocess_threads, batch_size_placeholder)
# Load the model
input_map = {'image_batch': image_batch, 'label_batch': label_batch, 'phase_train': phase_train_placeholder}
facenet.load_model(triplet_model, input_map=input_map)
# Get output tensor
embeddings = tf.get_default_graph().get_tensor_by_name("embeddings:0")
#
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
for n in range(len(all_paths)):
#print(n)
lfw_batch_size = len(all_paths[n])
if lfw_batch_size==0:
continue
emb_array = evaluate(sess, eval_enqueue_op, image_paths_placeholder, labels_placeholder, phase_train_placeholder,
batch_size_placeholder, control_placeholder, embeddings, label_batch, all_paths[n], lfw_batch_size, distance_metric)
if X[n,4:,X[n,0,:,1]+X[n,0,:,2]>0.5,0].shape[0]!=emb_array.shape[0]:
aa = 0
import pdb; pdb.set_trace()
#import pdb; pdb.set_trace()
X[n,4:,X[n,0,:,1]+X[n,0,:,2]>0.5,0] = emb_array
#import pdb; pdb.set_trace()
return X, Y
def main(_):
if FLAGS.csv_file_path:
if os.path.exists(FLAGS.csv_file_path) is False:
csv_dir = FLAGS.csv_file_path.rsplit('/', 1)[0]
if os.path.exists(csv_dir) is False:
os.makedirs(csv_dir)
if FLAGS.task_name == 'chalearn/age':
with open(FLAGS.csv_file_path, 'w') as f:
writer = csv.writer(f)
writer.writerow([
'Pruned rate', 'MAE', 'Acc', 'Epoch No.',
'Model size through inference (MB) (Shared part + task-specific part)',
'Shared part (MB)', 'Task specific part (MB)',
'Whole masks (MB)', 'Task specific masks (MB)',
'Task specific batch norm vars (MB)',
'Task specific biases (MB)'
])
else:
with open(FLAGS.csv_file_path, 'w') as f:
writer = csv.writer(f)
writer.writerow([
'Pruned rate', 'Acc', 'Epoch No.',
'Model size through inference (MB) (Shared part + task-specific part)',
'Shared part (MB)', 'Task specific part (MB)',
'Whole masks (MB)', 'Task specific masks (MB)',
'Task specific batch norm vars (MB)',
'Task specific biases (MB)'
])
args, unparsed = parse_arguments(sys.argv[1:])
FLAGS.filters_expand_ratio = math.sqrt(FLAGS.filters_expand_ratio)
FLAGS.history_filters_expand_ratios = [
math.sqrt(float(ratio))
for ratio in FLAGS.history_filters_expand_ratios
]
with tf.Graph().as_default():
with tf.Session() as sess:
if 'emotion' in FLAGS.task_name or 'chalearn' in FLAGS.task_name:
test_data_path = os.path.join(args.data_dir, 'val')
else:
test_data_path = os.path.join(args.data_dir, 'test')
test_set = utils.get_dataset(test_data_path)
# Get the paths for the corresponding images
image_list, label_list = facenet.get_image_paths_and_labels(
test_set)
image_paths_placeholder = tf.placeholder(tf.string,
shape=(None, 1),
name='image_paths')
labels_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='labels')
batch_size_placeholder = tf.placeholder(tf.int32,
name='batch_size')
control_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='control')
phase_train_placeholder = tf.placeholder(tf.bool,
name='phase_train')
nrof_preprocess_threads = 4
image_size = (args.image_size, args.image_size)
eval_input_queue = data_flow_ops.FIFOQueue(
capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1, ), (1, ), (1, )],
shared_name=None,
name=None)
eval_enqueue_op = eval_input_queue.enqueue_many(
[
image_paths_placeholder, labels_placeholder,
control_placeholder
],
name='eval_enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(
eval_input_queue, image_size, nrof_preprocess_threads,
batch_size_placeholder)
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
# Load the model
if os.path.isdir(args.model):
temp_record_file = os.path.join(args.model, 'temp_record.txt')
checkpoint_file = os.path.join(args.model, 'checkpoint')
if os.path.exists(temp_record_file) and os.path.exists(
checkpoint_file):
with open(temp_record_file) as json_file:
data = json.load(json_file)
max_acc = max(data, key=float)
epoch_no = data[max_acc]
ckpt_file = args.model + '/model-.ckpt-' + epoch_no
with open(checkpoint_file) as f:
context = f.read()
original_epoch = re.search("(\d)+", context).group()
context = context.replace(original_epoch, epoch_no)
with open(checkpoint_file, 'w') as f:
f.write(context)
if os.path.exists(os.path.join(args.model,
'copied')) is False:
os.makedirs(os.path.join(args.model, 'copied'))
copyfile(
temp_record_file,
os.path.join(args.model, 'copied', 'temp_record.txt'))
os.remove(temp_record_file)
elif os.path.exists(checkpoint_file):
ckpt = tf.train.get_checkpoint_state(args.model)
ckpt_file = ckpt.model_checkpoint_path
epoch_no = ckpt_file.rsplit('-', 1)[-1]
else:
print(
'No `temp_record.txt` or `checkpoint` in `{}`, you should pass args.model the file path, not the directory'
.format(args.model))
sys.exit(1)
else:
ckpt_file = args.model
epoch_no = ckpt_file.rsplit('-')[-1]
prelogits, _ = network.inference(
image_batch,
1.0,
phase_train=phase_train_placeholder,
bottleneck_layer_size=args.embedding_size,
weight_decay=0.0)
with tf.variable_scope('task_{}'.format(FLAGS.task_id)):
if FLAGS.task_name == 'chalearn/age':
logits = slim.fully_connected(prelogits,
100,
activation_fn=None,
scope='Logits',
reuse=False)
else:
logits = slim.fully_connected(prelogits,
len(test_set),
activation_fn=None,
scope='Logits',
reuse=False)
# Get output tensor
if FLAGS.task_name == 'chalearn/age':
softmax = tf.nn.softmax(logits=logits)
labels_range = tf.range(1.0, 101.0) # [1.0, ..., 100.0]
labels_matrix = tf.broadcast_to(
labels_range,
[args.test_batch_size, labels_range.shape[0]])
result_vector = tf.reduce_sum(softmax * labels_matrix, axis=1)
MAE_error_vector = tf.abs(result_vector -
tf.cast(label_batch, tf.float32))
MAE_avg_error = tf.reduce_mean(MAE_error_vector)
correct_prediction = tf.cast(
tf.equal(tf.argmax(logits, 1),
tf.cast(label_batch, tf.int64)), tf.float32)
accuracy = tf.reduce_mean(correct_prediction)
regularization_losses = tf.get_collection(
tf.GraphKeys.REGULARIZATION_LOSSES)
total_loss = tf.add_n([MAE_avg_error] + regularization_losses)
criterion = MAE_avg_error
else:
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(
labels=label_batch,
logits=logits,
name='cross_entropy_per_example')
cross_entropy_mean = tf.reduce_mean(cross_entropy)
correct_prediction = tf.cast(
tf.equal(tf.argmax(logits, 1),
tf.cast(label_batch, tf.int64)), tf.float32)
accuracy = tf.reduce_mean(correct_prediction)
regularization_losses = tf.get_collection(
tf.GraphKeys.REGULARIZATION_LOSSES)
total_loss = tf.add_n([cross_entropy_mean] +
regularization_losses)
criterion = cross_entropy_mean
init_fn = slim.assign_from_checkpoint_fn(ckpt_file,
tf.global_variables())
init_fn(sess)
pruned_ratio_relative_to_curr_task = 0.0
model_size = 0.0
if FLAGS.print_mem or FLAGS.print_mask_info:
masks = tf.get_collection('masks')
if FLAGS.print_mask_info:
if masks:
num_elems_in_each_task_op = {}
num_elems_in_tasks_in_masks_op = {
} # two dimentional dictionary
num_elems_in_masks_op = []
num_remain_elems_in_masks_op = []
for task_id in range(1, FLAGS.task_id + 1):
num_elems_in_each_task_op[task_id] = tf.constant(
0, dtype=tf.int32)
num_elems_in_tasks_in_masks_op[task_id] = {}
# Define graph
for i, mask in enumerate(masks):
num_elems_in_masks_op.append(tf.size(mask))
num_remain_elems_in_curr_mask = tf.size(mask)
for task_id in range(1, FLAGS.task_id + 1):
cnt = tf_count(mask, task_id)
num_elems_in_tasks_in_masks_op[task_id][
i] = cnt
num_elems_in_each_task_op[task_id] = tf.add(
num_elems_in_each_task_op[task_id], cnt)
num_remain_elems_in_curr_mask -= cnt
num_remain_elems_in_masks_op.append(
num_remain_elems_in_curr_mask)
num_elems_in_network_op = tf.add_n(
num_elems_in_masks_op)
print('Calculate pruning status ...')
# Doing operation
num_elems_in_masks = sess.run(num_elems_in_masks_op)
num_elems_in_each_task = sess.run(
num_elems_in_each_task_op)
num_elems_in_tasks_in_masks = sess.run(
num_elems_in_tasks_in_masks_op)
num_elems_in_network = sess.run(
num_elems_in_network_op)
num_remain_elems_in_masks = sess.run(
num_remain_elems_in_masks_op)
# Print out the result
print('Showing pruning status ...')
if FLAGS.verbose:
for i, mask in enumerate(masks):
print('Layer %s: ' % mask.op.name, end='')
for task_id in range(1, FLAGS.task_id + 1):
cnt = num_elems_in_tasks_in_masks[task_id][
i]
print('task_%d -> %d/%d (%.2f%%), ' %
(task_id, cnt, num_elems_in_masks[i],
100 * cnt / num_elems_in_masks[i]),
end='')
print('remain -> {:.2f}%'.format(
100 * num_remain_elems_in_masks[i] /
num_elems_in_masks[i]))
print('Num elems in network: {}'.format(
num_elems_in_network))
num_elems_of_usued_weights = num_elems_in_network
for task_id in range(1, FLAGS.task_id + 1):
print('Num elems in task_{}: {}'.format(
task_id, num_elems_in_each_task[task_id]))
print('Ratio of task_{} to all: {}'.format(
task_id, num_elems_in_each_task[task_id] /
num_elems_in_network))
num_elems_of_usued_weights -= num_elems_in_each_task[
task_id]
print('Num usued elems in all masks: {}'.format(
num_elems_of_usued_weights))
pruned_ratio_relative_to_all_elems = num_elems_of_usued_weights / num_elems_in_network
print('Ratio of usused_elem to all: {}'.format(
pruned_ratio_relative_to_all_elems))
pruned_ratio_relative_to_curr_task = num_elems_of_usued_weights / (
num_elems_of_usued_weights +
num_elems_in_each_task[FLAGS.task_id])
print('Pruning degree relative to task_{}: {:.3f}'.
format(FLAGS.task_id,
pruned_ratio_relative_to_curr_task))
if FLAGS.print_mem:
# Analyze param
start_time = time.time()
(MB_of_model_through_inference, MB_of_shared_variables,
MB_of_task_specific_variables, MB_of_whole_masks,
MB_of_task_specific_masks,
MB_of_task_specific_batch_norm_variables,
MB_of_task_specific_biases
) = model_analyzer.analyze_vars_for_current_task(
tf.model_variables(),
sess=sess,
task_id=FLAGS.task_id,
verbose=False)
duration = time.time() - start_time
print('duration time: {}'.format(duration))
if FLAGS.eval_once:
validate(
args, sess, image_list, label_list, eval_enqueue_op,
image_paths_placeholder, labels_placeholder,
control_placeholder, phase_train_placeholder,
batch_size_placeholder, total_loss, regularization_losses,
criterion, accuracy, args.use_fixed_image_standardization,
FLAGS.csv_file_path, pruned_ratio_relative_to_curr_task,
epoch_no, MB_of_model_through_inference,
MB_of_shared_variables, MB_of_task_specific_variables,
MB_of_whole_masks, MB_of_task_specific_masks,
MB_of_task_specific_batch_norm_variables,
MB_of_task_specific_biases)
return
def main(args):
network = importlib.import_module(args.model_def)
image_size = (args.image_size, args.image_size)
subdir = datetime.strftime(datatime.now(), '%Y%m%d-%H%M%S')
log_dir = os.path.join(os.path.expanduser(args.logs_base_dir), subdir)
if not os.path.isdir(log_dir):
os.makedirs(log_dir)
model_dir = os.path.join(os.path.expanduser(args.models_base_dir), subdir)
if not os.path.isdir(model_dir):
os.makedirs(model_dir)
stat_file_name = os.path.join(log_dir, 'stat.h5')
facenet.write_arguments_to_file(args, os.path.join(log_dir, 'arguments.txt'))
src_path,_ = os.path.split(os.path.realpath(__file__))
facenet.store_revision_info(src_path, log_dir, ' '.join(sys.argv))
np.random.seed(seed=args.seed)
random.seed(args.seed)
dataset = facenet.get_dataset(args.data_dir)
if args.validation_set_split_ratio > 0.0:
train_set, val_set = facenet.split_dataset(dataset, args.validation_set_split_ratio, args.min_nrof_val_images_per_class,
'SPLIT_IMAGES')
else:
train_set, val_set = dataset, []
num_classes = len(train_set)
print('Model directory: %s' % model_dir)
print('Log directory: %s' % log_dir)
pretrained_model = None
if args.pretrained_model:
pretrained_model = os.path.expanduser(args.pretrained_model)
print('Pre-trained model: %s' % pretrained_model)
if args.lfw_dir:
print('LFW directory: %s' % args.lfw_dir)
pairs = lfw.read_pairs(os.path.expanduser(args.lfw_pairs))
lfw_paths, actual_issame = lfw.get_paths(os.path.expanduser(args.lfw_dir), pairs)
with tf.Graph().as_default():
tf.set_random_state(args.seed)
global_step = tf.Variable(0, trainable=False)
image_list, label_list = facenet.get_image_paths_and_labels(train_set)
assert len(image_list) > 0, 'The training set should not be empty'
val_image_list, val_label_list = facenet.get_image_paths_and_labels(val_set)
labels = ops.convert_to_tensor(label_list, dtype=tf.int32)
range_size = array_ops.shape(labels)[0]
index_q = tf.train.range_input_producer(range_size, num_epochs=None, shuffle=True, seed=None, capacity=32)
index_deq_op = index_q.dequeue_many(args.batch_size * args.epoch_size, 'index_dequeue')
learning_rate_placeholder = tf.placeholder(tf.float32, name='learning_rate')
batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size')
phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train')
image_paths_placeholder = tf.placeholder(tf.string, shape=(None, 1), name='image_paths')
labels_placeholder = tf.placeholder(tf.int32, shape=(None, 1), name='labels')
control_placeholder = tf.placeholder(tf.int32, shape=(None, 1), name='control')
num_preprocess_threads = 4
input_q = data_flow_ops.FIFOQueue(capacity=2000000, shared_name=None, name=None,
dtype=[tf.string, tf.int32, tf.int32],
shapes=[(1, ), (1, ), (1, )])
enq_op = input_q.enqueue_many([image_paths_placeholder, labels_placeholder, control_placeholder], name='enq_op')
image_batch, label_batch = facenet.create_input_pipeline(input_q, image_size, num_preprocess_threads, batch_size_placeholder)
image_batch = tf.identity(image_batch, 'image_batch')
image_batch = tf.identity(image_batch, 'input')
label_batch = tf.identity(label_batch, 'label_batch')
print('Number of classes in training set: %d' % nrof_classes)
print('Number of examples in training set: %d' % len(image_list))
print('Number of classes in validation set: %d' % len(val_set))
print('Number of examples in validation set: %d' % len(val_image_list))
print('Building training graph')
prelogits, _ = network.inference(image_batch, args.keep_probability, phase_train=phase_train_placeholder,
bottleneck_layer_size=args.embedding_size, weight_decay=args.weight_decay)
logits = slim.fully_connected(prelogits, num_classes, activation_fn=None, reuse=False, scope='Logits',
weights_initializer=slim.initializers.xavier_initializer(),
weights_regularizer=slim.l2_regularizer(args.weight_decay))
embeddings = tf.nn.l2_normalize(prelogits, 1, 1e-10, name='embeddings')
eps = 1e-4
prelogits_norm = tf.reduce_mean(tf.norm(tf.abs(prelogits) + eps, ord=args.prelogits_norm_p, axis=1))
tf.add_to_collection(tf.GraphKeys.REGULARIZATION_LOSSES, prelogits_norm * args.prelogits_norm_loss_factor)
prelogits_center_loss, _ = facenet.center_loss(prelogits, label_batch, args.center_loss_alfa, num_classes)
tf.add_to_collection(tf.GraphKeys.REGULARIZATION_LOSSES, prelogits_center_loss * args.center_loss_factor)
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=label_batch, logits=logits, name='cross_entropy_per_example')
cross_entropy_mean = tf.reduce_mean(cross_entropy, name='cross_entropy')
tf.add_to_collection('losses', cross_entropy_mean)
regularization_loss = tf.get_collection(tf.GraphKeys.REGULARIZATION_LOSSES)
total_loss = tf.add_n([cross_entropy_mean] + regularization_loss, name='total_loss')
correct_prediction = tf.cast(tf.equal(tf.argmax(logits, 1), tf.cast(label_batch, tf.int64)), tf.float32)
accuracy = tf.reduce_mean(correct_prediction)
learning_rate = tf.train.exponential_decay(learning_rate_placeholder, global_step,
args.learning_rate_decay_epochs * args.epoch_size,
args.learning_rate_decay_factor, staircase=True,)
tf.summary.scalar('learning_rate', learning_rate)
train_op = facenet.train(total_loss, global_step, args.optimizer, learning_rate,
args.moving_average_decay, tf.global_variables(), args.log_histograms)
saver = tf.train.Saver(tf.trainable_variables(), max_to_keep=3)
summary_op = tf.summary.merge_all()
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=args.gpu_memory_fraction)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options, log_device_placement=False))
sess.run(tf.global_variables_initializer())
sess.run(tf.local_variables_initializer())
summary_writer = tf.summary.FileWriter(log_dir, sess.graph)
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
with sess.as_default():
if pretrained_model:
print('Restoring pretrained model: %s' % pretrained_model)
saver.restore(sess, pretrained_model)
print('Running training')
num_steps = args.max_nrof_epochs * args.epoch_size
num_val_samples = int(math.ceil(args.max_nrof_epochs / args.validate_every_n_epochs))
stat = {
'loss': np.zeros((num_steps,), np.float32),
'center_loss': np.zeros((num_steps,), np.float32),
'reg_loss': np.zeros((num_steps,), np.float32),
'xent_loss': np.zeros((num_steps,), np.float32),
'prelogits_norm': np.zeros((num_steps,), np.float32),
'accuracy': np.zeros((num_steps,), np.float32),
'val_loss': np.zeros((num_val_samples,), np.float32),
'val_xent_loss': np.zeros((num_val_samples,), np.float32),
'val_accuracy': np.zeros((num_val_samples,), np.float32),
'lfw_accuracy': np.zeros((args.max_nrof_epochs,), np.float32),
'lfw_valrate': np.zeros((args.max_nrof_epochs,), np.float32),
'learning_rate': np.zeros((args.max_nrof_epochs,), np.float32),
'time_train': np.zeros((args.max_nrof_epochs,), np.float32),
'time_validate': np.zeros((args.max_nrof_epochs,), np.float32),
'time_evaluate': np.zeros((args.max_nrof_epochs,), np.float32),
'prelogits_hist': np.zeros((args.max_nrof_epochs, 1000), np.float32),
}
for epoch in range(1, args.max_nrof_epochs + 1):
step = sess.run(global_step, feed_dict=None)
t = time.time()
cont = train(args, sess, epoch, image_list, label_list, index_dequeue_op, enqueue_op,
image_paths_placeholder, labels_placeholder, learning_rate_placeholder, phase_train_placeholder,
batch_size_placeholder, control_placeholder, global_step, total_loss, train_op, summary_op,
summary_writer, regularization_losses, args.learning_rate_schedule_file, stat, cross_entropy_mean,
accuracy, learning_rate, prelogits, prelogits_center_loss, args.random_rotate, args.random_crop,
args.random_flip, prelogits_norm, args.prelogits_hist_max, args.use_fixed_image_standardization)
stat['time_train'][epoch - 1] = time.time() - t
if not cont:
break
t = time.time()
if len(val_image_list)>0 and ((epoch-1) % args.validate_every_n_epochs == args.validate_every_n_epochs-1 or
epoch==args.max_nrof_epochs):
validate(args, sess, epoch, val_image_list, val_label_list, enqueue_op, image_paths_placeholder,
labels_placeholder, control_placeholder, phase_train_placeholder, batch_size_placeholder, stat,
total_loss, regularization_losses, cross_entropy_mean, accuracy, args.validate_every_n_epochs,
args.use_fixed_image_standardization)
stat['time_validate'][epoch - 1] = time.time() - t
save_variables_and_metagraph(sess, saver, summary_writer, model_dir, subdir, epoch)
t = time.time()
if args.lfw_dir:
evaluate(sess, enqueue_op, image_paths_placeholder, labels_placeholder, phase_train_placeholder,
batch_size_placeholder, control_placeholder, embeddings, label_batch, lfw_paths, actual_issame,
args.lfw_batch_size, args.lfw_nrof_folds, log_dir, step, summary_writer, stat, epoch,
args.lfw_distance_metric, args.lfw_subtract_mean, args.lfw_use_flipped_images,
args.use_fixed_image_standardization)
stat['time_evaluate'][epoch - 1] = time.time() - t
print('Saving statistics')
with h5py.File(stat_file_name, 'w') as f:
for key, value in stat.items():
f.create_dataset(key, data=value)
return model_dir
def main(args):
# 模型，定义在inception_resnet_v1 V2里（）， --model_def models.inception_resnet_v1
network = importlib.import_module(args.model_def)
image_size = (args.image_size, args.image_size)
subdir = datetime.strftime(datetime.now(), '%Y%m%d-%H%M%S')
log_dir = os.path.join(os.path.expanduser(args.logs_base_dir), subdir)
if not os.path.isdir(log_dir): # Create the log directory if it doesn't exist
os.makedirs(log_dir)
model_dir = os.path.join(os.path.expanduser(args.models_base_dir), subdir)
if not os.path.isdir(model_dir): # Create the model directory if it doesn't exist
os.makedirs(model_dir)
stat_file_name = os.path.join(log_dir, 'stat.h5')
# Write arguments to a text file
facenet.write_arguments_to_file(args, os.path.join(log_dir, 'arguments.txt'))
# Store some git revision info in a text file in the log directory
src_path,_ = os.path.split(os.path.realpath(__file__))
facenet.store_revision_info(src_path, log_dir, ' '.join(sys.argv))
np.random.seed(seed=args.seed)
random.seed(args.seed)
dataset = facenet.get_dataset(args.data_dir)
if args.filter_filename:
dataset = filter_dataset(dataset, os.path.expanduser(args.filter_filename),
args.filter_percentile, args.filter_min_nrof_images_per_class)
if args.validation_set_split_ratio>0.0:
train_set, val_set = facenet.split_dataset(dataset, args.validation_set_split_ratio, args.min_nrof_val_images_per_class, 'SPLIT_IMAGES')
else:
train_set, val_set = dataset, []
nrof_classes = len(train_set)
print('Model directory: %s' % model_dir)
print('Log directory: %s' % log_dir)
pretrained_model = None
if args.pretrained_model:
pretrained_model = os.path.expanduser(args.pretrained_model)
print('Pre-trained model: %s' % pretrained_model)
if args.lfw_dir:
print('LFW directory: %s' % args.lfw_dir)
# Read the file containing the pairs used for testing
pairs = lfw.read_pairs(os.path.expanduser(args.lfw_pairs))
# Get the paths for the corresponding images
lfw_paths, actual_issame = lfw.get_paths(os.path.expanduser(args.lfw_dir), pairs)
with tf.Graph().as_default():
tf.set_random_seed(args.seed)
global_step = tf.Variable(0, trainable=False)
# Get a list of image paths and their labels
# 训练数据
image_list, label_list = facenet.get_image_paths_and_labels(train_set)
assert len(image_list)>0, 'The training set should not be empty'
# 测试数据
val_image_list, val_label_list = facenet.get_image_paths_and_labels(val_set)
# Create a queue that produces indices into the image_list and label_list
# tf.convert_to_tensor用于将不同数据变成张量：比如可以让数组变成张量、也可以让列表变成张量。
labels = ops.convert_to_tensor(label_list, dtype=tf.int32)
range_size = array_ops.shape(labels)[0]
# 多线程读取数据，shuffle=True表示不是按顺序存储，可以随机获取，并一直循环。
# https://blog.csdn.net/lyg5623/article/details/69387917
index_queue = tf.train.range_input_producer(range_size, num_epochs=None,
shuffle=True, seed=None, capacity=32)
# epoch 大数据时迭代完一轮时次数，少量数据应该epoch = 全部数据个数/batch
index_dequeue_op = index_queue.dequeue_many(args.batch_size*args.epoch_size, 'index_dequeue')
learning_rate_placeholder = tf.placeholder(tf.float32, name='learning_rate')
batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size')
phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train')
image_paths_placeholder = tf.placeholder(tf.string, shape=(None,1), name='image_paths')
labels_placeholder = tf.placeholder(tf.int32, shape=(None,1), name='labels')
control_placeholder = tf.placeholder(tf.int32, shape=(None,1), name='control')
nrof_preprocess_threads = 4
input_queue = data_flow_ops.FIFOQueue(capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1,), (1,), (1,)],
shared_name=None, name=None)
enqueue_op = input_queue.enqueue_many([image_paths_placeholder, labels_placeholder, control_placeholder], name='enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(input_queue, image_size, nrof_preprocess_threads, batch_size_placeholder)
image_batch = tf.identity(image_batch, 'image_batch')
image_batch = tf.identity(image_batch, 'input')
label_batch = tf.identity(label_batch, 'label_batch')
print('Number of classes in training set: %d' % nrof_classes)
print('Number of examples in training set: %d' % len(image_list))
print('Number of classes in validation set: %d' % len(val_set))
print('Number of examples in validation set: %d' % len(val_image_list))
print('Building training graph')
# Build the inference graph
prelogits, _ = network.inference(image_batch, args.keep_probability,
phase_train=phase_train_placeholder, bottleneck_layer_size=args.embedding_size,
weight_decay=args.weight_decay)
# 因为模型输出的(bottleneck_layer_size）没有计算最后一层(映射到图片类型)，这里计算最后一层
logits = slim.fully_connected(prelogits, len(train_set), activation_fn=None,
weights_initializer=slim.initializers.xavier_initializer(),
weights_regularizer=slim.l2_regularizer(args.weight_decay),
scope='Logits', reuse=False)
# 按行进行泛化，行的平方求和再求平方根，得到的值按行除每个行的元素，对深度层面泛化? interface里最后一层输出为128个节点，slim.fully_connected(net, bottleneck_layer_size, activation_fn=None,
#https://blog.csdn.net/abiggg/article/details/79368982
embeddings = tf.nn.l2_normalize(prelogits, 1, 1e-10, name='embeddings')
# 计算loss函数，当然还有其它训练参数也会加到这里来，通过比训练过程中一个weight加到正则化参数里来tf.add_to_collection(tf.GraphKeys.REGULARIZATION_LOSSES, weight)
# 模型中最后会把这个加到优化的loss中来。
#L= L_softmax + λL_cneter = Softmax(W_i + b_yj) + λ1/2||f(x_i) - c_yj ||_2^2
# Norm for the prelogits
eps = 1e-4
prelogits_norm = tf.reduce_mean(tf.norm(tf.abs(prelogits)+eps, ord=args.prelogits_norm_p, axis=1))
# 模型中最后输出(bottleneck_layer_size每个类型的输出值的个数)的平均值加到正则化loss中，但prelogits_norm_loss_factor貌似为0
tf.add_to_collection(tf.GraphKeys.REGULARIZATION_LOSSES, prelogits_norm * args.prelogits_norm_loss_factor)
# 计算中心损失及增加的正则化loss中
# Add center loss
prelogits_center_loss, _ = facenet.center_loss(prelogits, label_batch, args.center_loss_alfa, nrof_classes)
tf.add_to_collection(tf.GraphKeys.REGULARIZATION_LOSSES, prelogits_center_loss * args.center_loss_factor)
learning_rate = tf.train.exponential_decay(learning_rate_placeholder, global_step,
args.learning_rate_decay_epochs*args.epoch_size, args.learning_rate_decay_factor, staircase=True)
tf.summary.scalar('learning_rate', learning_rate)
# Calculate the average cross entropy loss across the batch
# 计算预测损失，和上面框架的Softmax(W_i + b_yj)
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(
labels=label_batch, logits=logits, name='cross_entropy_per_example')
cross_entropy_mean = tf.reduce_mean(cross_entropy, name='cross_entropy')
# 预测损失平均值加到losses变量中
tf.add_to_collection('losses', cross_entropy_mean)
correct_prediction = tf.cast(tf.equal(tf.argmax(logits, 1), tf.cast(label_batch, tf.int64)), tf.float32)
accuracy = tf.reduce_mean(correct_prediction)
#计算总损失，cross_entropy_mean + 前面增加的一些正则化损失(包括模型中增加的)，通过tf.GraphKeys.REGULARIZATION_LOSSES获取出来
# Calculate the total losses
regularization_losses = tf.get_collection(tf.GraphKeys.REGULARIZATION_LOSSES)
total_loss = tf.add_n([cross_entropy_mean] + regularization_losses, name='total_loss')
# Build a Graph that trains the model with one batch of examples and updates the model parameters
train_op = facenet.train(total_loss, global_step, args.optimizer,
learning_rate, args.moving_average_decay, tf.global_variables(), args.log_histograms)
# Create a saver
saver = tf.train.Saver(tf.trainable_variables(), max_to_keep=3)
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.summary.merge_all()
# Start running operations on the Graph.
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=args.gpu_memory_fraction)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options, log_device_placement=False))
sess.run(tf.global_variables_initializer())
sess.run(tf.local_variables_initializer())
summary_writer = tf.summary.FileWriter(log_dir, sess.graph)
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
with sess.as_default():
if pretrained_model:
print('Restoring pretrained model: %s' % pretrained_model)
saver.restore(sess, pretrained_model)
# Training and validation loop
print('Running training')
nrof_steps = args.max_nrof_epochs*args.epoch_size
nrof_val_samples = int(math.ceil(args.max_nrof_epochs / args.validate_every_n_epochs)) # Validate every validate_every_n_epochs as well as in the last epoch
stat = {
'loss': np.zeros((nrof_steps,), np.float32),
'center_loss': np.zeros((nrof_steps,), np.float32),
'reg_loss': np.zeros((nrof_steps,), np.float32),
'xent_loss': np.zeros((nrof_steps,), np.float32),
'prelogits_norm': np.zeros((nrof_steps,), np.float32),
'accuracy': np.zeros((nrof_steps,), np.float32),
'val_loss': np.zeros((nrof_val_samples,), np.float32),
'val_xent_loss': np.zeros((nrof_val_samples,), np.float32),
'val_accuracy': np.zeros((nrof_val_samples,), np.float32),
'lfw_accuracy': np.zeros((args.max_nrof_epochs,), np.float32),
'lfw_valrate': np.zeros((args.max_nrof_epochs,), np.float32),
'learning_rate': np.zeros((args.max_nrof_epochs,), np.float32),
'time_train': np.zeros((args.max_nrof_epochs,), np.float32),
'time_validate': np.zeros((args.max_nrof_epochs,), np.float32),
'time_evaluate': np.zeros((args.max_nrof_epochs,), np.float32),
'prelogits_hist': np.zeros((args.max_nrof_epochs, 1000), np.float32),
}
for epoch in range(1,args.max_nrof_epochs+1):
step = sess.run(global_step, feed_dict=None)
# Train for one epoch
t = time.time()
# 训练模型
cont = train(args, sess, epoch, image_list, label_list, index_dequeue_op, enqueue_op, image_paths_placeholder, labels_placeholder,
learning_rate_placeholder, phase_train_placeholder, batch_size_placeholder, control_placeholder, global_step,
total_loss, train_op, summary_op, summary_writer, regularization_losses, args.learning_rate_schedule_file,
stat, cross_entropy_mean, accuracy, learning_rate,
prelogits, prelogits_center_loss, args.random_rotate, args.random_crop, args.random_flip, prelogits_norm, args.prelogits_hist_max, args.use_fixed_image_standardization)
stat['time_train'][epoch-1] = time.time() - t
if not cont:
break
# 在测试数据上计算正确率
t = time.time()
if len(val_image_list)>0 and ((epoch-1) % args.validate_every_n_epochs == args.validate_every_n_epochs-1 or epoch==args.max_nrof_epochs):
validate(args, sess, epoch, val_image_list, val_label_list, enqueue_op, image_paths_placeholder, labels_placeholder, control_placeholder,
phase_train_placeholder, batch_size_placeholder,
stat, total_loss, regularization_losses, cross_entropy_mean, accuracy, args.validate_every_n_epochs, args.use_fixed_image_standardization)
stat['time_validate'][epoch-1] = time.time() - t
# Save variables and the metagraph if it doesn't exist already
save_variables_and_metagraph(sess, saver, summary_writer, model_dir, subdir, epoch)
# Evaluate on LFW
t = time.time()
if args.lfw_dir:
evaluate(sess, enqueue_op, image_paths_placeholder, labels_placeholder, phase_train_placeholder, batch_size_placeholder, control_placeholder,
embeddings, label_batch, lfw_paths, actual_issame, args.lfw_batch_size, args.lfw_nrof_folds, log_dir, step, summary_writer, stat, epoch,
args.lfw_distance_metric, args.lfw_subtract_mean, args.lfw_use_flipped_images, args.use_fixed_image_standardization)
stat['time_evaluate'][epoch-1] = time.time() - t
print('Saving statistics')
with h5py.File(stat_file_name, 'w') as f:
for key, value in stat.items():
f.create_dataset(key, data=value)
return model_dir
def main(args):
network = importlib.import_module(args.model_def)
image_size = (args.image_size, args.image_size)
np.random.seed(seed=args.seed)
pretrained_model = None
pretrained_model = os.path.expanduser(args.pretrained_model)
print('Pre-trained model: %s' % pretrained_model)
print('LFW directory: %s' % args.lfw_dir)
pairs = lfw.read_pairs(os.path.expanduser(args.lfw_pairs))
lfw_paths, actual_issame = lfw.get_paths(os.path.expanduser(args.lfw_dir),
pairs)
with tf.Graph().as_default():
tf.set_random_seed(args.seed)
global_step = tf.Variable(0, trainable=False)
batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size')
phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train')
image_paths_placeholder = tf.placeholder(tf.string,
shape=(None, 1),
name='image_paths')
labels_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='labels')
control_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='control')
nrof_preprocess_threads = 4
input_queue = data_flow_ops.FIFOQueue(
capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1, ), (1, ), (1, )],
shared_name=None,
name=None)
enqueue_op = input_queue.enqueue_many(
[image_paths_placeholder, labels_placeholder, control_placeholder],
name='enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(
input_queue, image_size, nrof_preprocess_threads,
batch_size_placeholder)
image_batch = tf.identity(image_batch, 'image_batch')
image_batch = tf.identity(image_batch, 'input')
label_batch = tf.identity(label_batch, 'label_batch')
prelogits, _ = network.inference(
image_batch,
args.keep_probability,
phase_train=phase_train_placeholder,
bottleneck_layer_size=args.embedding_size,
weight_decay=args.weight_decay)
embeddings = tf.nn.l2_normalize(prelogits, 1, 1e-10, name='embeddings')
# Start running operations on the Graph.
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=args.
gpu_memory_fraction) #设置GPU显存占用率
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options,
log_device_placement=False))
sess.run(tf.global_variables_initializer())
sess.run(tf.local_variables_initializer())
# tensorflow中的多进程管理,下面两条语句通常配套使用
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
with sess.as_default():
if pretrained_model:
print('Restoring pretrained model: %s' % pretrained_model)
facenet.load_model(pretrained_model)
# for epoch in range(1,args.max_nrof_epochs+1):
for epoch in range(1, 2):
step = sess.run(global_step, feed_dict=None)
# Evaluate on LFW
t = time.time()
if args.lfw_dir:
evaluate(sess, enqueue_op, image_paths_placeholder,
labels_placeholder, phase_train_placeholder,
batch_size_placeholder, control_placeholder,
embeddings, label_batch, lfw_paths, actual_issame,
args.lfw_batch_size, args.lfw_nrof_folds, step,
epoch, args.lfw_distance_metric,
args.lfw_subtract_mean,
args.lfw_use_flipped_images,
args.use_fixed_image_standardization)
def PersonCompare(info_queue, msg_q):
image_size = (160, 160)
total_dist = 0.0000
with tf.Graph().as_default():
with tf.Session() as sess:
#image_dir1 = os.path.expanduser(args.first_dir)
#image_dir2 = os.path.expanduser(args.second_dir)
image_paths_placeholder = tf.placeholder(tf.string,
shape=(None, 1),
name='image_paths')
labels_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='labels')
batch_size_placeholder = tf.placeholder(tf.int32,
name='batch_size')
control_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='control')
phase_train_placeholder = tf.placeholder(tf.bool,
name='phase_train')
nrof_preprocess_threads = 4
#image_size = (image_size, image_size)
eval_input_queue = data_flow_ops.FIFOQueue(
capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1, ), (1, ), (1, )],
shared_name=None,
name=None)
eval_enqueue_op = eval_input_queue.enqueue_many(
[
image_paths_placeholder, labels_placeholder,
control_placeholder
],
name='eval_enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(
eval_input_queue, image_size, nrof_preprocess_threads,
batch_size_placeholder)
# Load the model
input_map = {
'image_batch': image_batch,
'label_batch': label_batch,
'phase_train': phase_train_placeholder
}
print('=== model: ')
print(model)
print('===input_map: ')
print(input_map)
facenet.load_model(model, input_map=input_map)
images_placeholder = tf.get_default_graph().get_tensor_by_name(
"input:0")
# Get output tensor
embeddings = tf.get_default_graph().get_tensor_by_name(
"embeddings:0")
embedding_size = embeddings.get_shape()[1]
print('facenet embedding模型建立完毕')
image_num = 0
pass_num = 0
while 1:
input_img = info_queue.get()
you = "who ?"
last_dist = 2
#print(input_img)
#print("compare dir: %s" % compare_dir)
for parent, dirnames, filenames in os.walk(compare_dir):
#print(dirnames)
for username in dirnames:
userdir = os.path.join(compare_dir, username)
img_files = os.path.join(userdir,
"*" + image_extension)
img_name_list = glob.glob(img_files)
for compare_img in img_name_list:
scaled_reshape = []
image1 = scipy.misc.imread(input_img, mode='RGB')
image1 = cv2.resize(image1,
image_size,
interpolation=cv2.INTER_CUBIC)
image1 = facenet.prewhiten(image1)
scaled_reshape.append(
image1.reshape(-1, 160, 160, 3))
emb_array1 = np.zeros((1, embedding_size))
emb_array1[0, :] = sess.run(
embeddings,
feed_dict={
images_placeholder: scaled_reshape[0],
phase_train_placeholder: False
})[0]
image2 = scipy.misc.imread(compare_img, mode='RGB')
image2 = cv2.resize(image2,
image_size,
interpolation=cv2.INTER_CUBIC)
image2 = facenet.prewhiten(image2)
scaled_reshape.append(
image2.reshape(-1, 160, 160, 3))
emb_array2 = np.zeros((1, embedding_size))
emb_array2[0, :] = sess.run(
embeddings,
feed_dict={
images_placeholder: scaled_reshape[1],
phase_train_placeholder: False
})[0]
dist = np.sqrt(
np.sum(np.square(emb_array1[0] -
emb_array2[0])))
#print("%s：%f "% (username,dist))
total_dist += dist
#if dist < 0.9:
#print("pass num += 1")
#pass_num += 1
image_num += 1
#print(" 第%d组照片是同一个人 "%num)
if image_num == 0:
#pass_num = 0
total_dist = 0
continue
#pass_rate = pass_num/image_num
#print("%s 通过率: %f " % (username,pass_rate))
#pass_num = 0
mean_dist = total_dist / image_num
#print("---- %s mean dist: %f" % (username,mean_dist))
if mean_dist < last_dist:
last_dist = mean_dist
you = username
image_num = 0
total_dist = 0
#if pass_rate >= 0.8 :
#print("=========== 你是: %s\n" % username)
#break
print("=========== 你是: %s\n" % you)
break
if not msg_q.empty():
if msg_q.get() == "complate_msg":
break
def train_facenet(config_):
# display the model parameters
config_.display()
# Get the image_size
image_size = (config_.im_height, config_.im_width)
preprocess(config_)
# Create dirs to save the logs and checkpoints
subdir = datetime.strftime(datetime.now(), '%Y%m%d-%H%M%S')
log_dir = os.path.join(os.path.expanduser(config_.logs_base_dir), subdir)
if not os.path.isdir(
log_dir): # Create the log directory if it doesn't exist
os.makedirs(log_dir)
model_dir = os.path.join(os.path.expanduser(config_.models_base_dir),
subdir)
if not os.path.isdir(
model_dir): # Create the model directory if it doesn't exist
os.makedirs(model_dir)
# random seed for reproducability
np.random.seed(seed=config_.seed)
random.seed(config_.seed)
# get the training dataset, the dir should have structure
# person1
# person1_image_1
# person1_image_2
# .
# .
# person 2
# person2_image_1
# .
# .
dataset = facenet.get_dataset(config_.data_dir)
# get train and test dataset
if config_.validation_set_split_ratio > 0.0:
train_set, val_set = facenet.split_dataset(
dataset, config_.validation_set_split_ratio,
config_.min_nrof_val_images_per_class, 'SPLIT_IMAGES')
else:
train_set, val_set = dataset, []
nrof_classes = len(train_set)
print('Model directory: %s' % model_dir)
print('Log directory: %s' % log_dir)
pretrained_model = None
if config_.pretrained_model:
pretrained_model = os.path.expanduser(config_.pretrained_model)
print('Pre-trained model: %s' % pretrained_model)
# get the image pairs to test the quality of embedding
if config_.lfw_dir:
print('LFW directory: %s' % config_.lfw_dir)
# Read the file containing the pairs used for testing
pairs = lfw.read_pairs(os.path.expanduser(config_.lfw_pairs))
# Get the paths for the corresponding images
lfw_paths, actual_issame = lfw.get_paths(
os.path.expanduser(config_.lfw_dir), pairs)
with tf.Graph().as_default():
tf.set_random_seed(config_.seed)
global_step = tf.Variable(0, trainable=False)
# Get a list of image paths and their labels
image_list, label_list = facenet.get_image_paths_and_labels(train_set)
assert len(image_list) > 0, 'The training set should not be empty'
val_image_list, val_label_list = facenet.get_image_paths_and_labels(
val_set)
# Create a queue that produces indices into the image_list and label_list
labels = ops.convert_to_tensor(label_list, dtype=tf.int32)
range_size = array_ops.shape(labels)[0]
index_queue = tf.train.range_input_producer(range_size,
num_epochs=None,
shuffle=True,
seed=None,
capacity=32)
index_dequeue_op = index_queue.dequeue_many(
config_.batch_size * config_.epoch_size, 'index_dequeue')
learning_rate_placeholder = tf.placeholder(tf.float32,
name='learning_rate')
batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size')
phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train')
image_paths_placeholder = tf.placeholder(tf.string,
shape=(None, 1),
name='image_paths')
labels_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='labels')
control_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='control')
nrof_preprocess_threads = 4
input_queue = data_flow_ops.FIFOQueue(
capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1, ), (1, ), (1, )],
shared_name=None,
name=None)
enqueue_op = input_queue.enqueue_many(
[image_paths_placeholder, labels_placeholder, control_placeholder],
name='enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(
input_queue, image_size, nrof_preprocess_threads,
batch_size_placeholder)
image_batch = tf.identity(image_batch, 'image_batch')
image_batch = tf.identity(image_batch, 'input')
label_batch = tf.identity(label_batch, 'label_batch')
print('Number of classes in training set: %d' % nrof_classes)
print('Number of examples in training set: %d' % len(image_list))
print('Number of classes in validation set: %d' % len(val_set))
print('Number of examples in validation set: %d' % len(val_image_list))
print('Building training graph')
# Build the inference graph
prelogits, _ = network.inference(
image_batch,
config_.keep_probability,
phase_train=phase_train_placeholder,
bottleneck_layer_size=config_.embedding_size,
weight_decay=config_.weight_decay)
# logits needed for training, we can ignore this in testing phase
logits = slim.fully_connected(
prelogits,
len(train_set),
activation_fn=None,
weights_initializer=slim.initializers.xavier_initializer(),
weights_regularizer=slim.l2_regularizer(config_.weight_decay),
scope='Logits',
reuse=False)
# learned embeddings, will be used to assess the similarity between two images
embeddings = tf.nn.l2_normalize(prelogits, 1, 1e-10, name='embeddings')
# Norm for the prelogits
eps = 1e-4
prelogits_norm = tf.reduce_mean(
tf.norm(tf.abs(prelogits) + eps,
ord=config_.prelogits_norm_p,
axis=1))
tf.add_to_collection(
tf.GraphKeys.REGULARIZATION_LOSSES,
prelogits_norm * config_.prelogits_norm_loss_factor)
# Add center loss
prelogits_center_loss, _ = facenet.center_loss(
prelogits, label_batch, config_.center_loss_alfa, nrof_classes)
tf.add_to_collection(
tf.GraphKeys.REGULARIZATION_LOSSES,
prelogits_center_loss * config_.center_loss_factor)
learning_rate = tf.train.exponential_decay(
learning_rate_placeholder,
global_step,
config_.learning_rate_decay_epochs * config_.epoch_size,
config_.learning_rate_decay_factor,
staircase=True)
tf.summary.scalar('learning_rate', learning_rate)
# Calculate the average cross entropy loss across the batch
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(
labels=label_batch,
logits=logits,
name='cross_entropy_per_example')
cross_entropy_mean = tf.reduce_mean(cross_entropy,
name='cross_entropy')
tf.add_to_collection('losses', cross_entropy_mean)
correct_prediction = tf.cast(
tf.equal(tf.argmax(logits, 1), tf.cast(label_batch, tf.int64)),
tf.float32)
accuracy = tf.reduce_mean(correct_prediction)
# Calculate the total losses
regularization_losses = tf.get_collection(
tf.GraphKeys.REGULARIZATION_LOSSES)
total_loss = tf.add_n([cross_entropy_mean] + regularization_losses,
name='total_loss')
# Build a Graph that trains the model with one batch of examples and updates the model parameters
train_op = facenet.train(total_loss, global_step, config_.optimizer,
learning_rate, config_.moving_average_decay,
tf.global_variables(), config_.log_histograms)
# Create a saver
all_vars = tf.trainable_variables()
var_to_restore = [
v for v in all_vars if not v.name.startswith('Logits')
]
saver = tf.train.Saver(var_to_restore,
max_to_keep=5,
keep_checkpoint_every_n_hours=1.0)
#saver = tf.train.Saver(tf.trainable_variables(), max_to_keep=3)
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.summary.merge_all()
# Start running operations on the Graph.
gpu_options = tf.GPUOptions(
per_process_gpu_memory_fraction=config_.gpu_memory_fraction)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options,
log_device_placement=False))
sess.run(tf.global_variables_initializer())
sess.run(tf.local_variables_initializer())
summary_writer = tf.summary.FileWriter(log_dir, sess.graph)
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
with sess.as_default():
if pretrained_model:
try:
print('Restoring pretrained model: %s' % pretrained_model)
#saver.restore(sess, pretrained_model)
facenet.load_model(pretrained_model)
except:
print(
'Error loading model(maybe checkpoint path is wrong).. training from scratch'
)
# Training and validation loop
print('Running training')
nrof_steps = config_.max_nrof_epochs * config_.epoch_size
nrof_val_samples = int(
math.ceil(config_.max_nrof_epochs /
config_.validate_every_n_epochs)
) # Validate every validate_every_n_epochs as well as in the last epoch
stat = {
'loss':
np.zeros((nrof_steps, ), np.float32),
'center_loss':
np.zeros((nrof_steps, ), np.float32),
'reg_loss':
np.zeros((nrof_steps, ), np.float32),
'xent_loss':
np.zeros((nrof_steps, ), np.float32),
'prelogits_norm':
np.zeros((nrof_steps, ), np.float32),
'accuracy':
np.zeros((nrof_steps, ), np.float32),
'val_loss':
np.zeros((nrof_val_samples, ), np.float32),
'val_xent_loss':
np.zeros((nrof_val_samples, ), np.float32),
'val_accuracy':
np.zeros((nrof_val_samples, ), np.float32),
'lfw_accuracy':
np.zeros((config_.max_nrof_epochs, ), np.float32),
'lfw_valrate':
np.zeros((config_.max_nrof_epochs, ), np.float32),
'learning_rate':
np.zeros((config_.max_nrof_epochs, ), np.float32),
'time_train':
np.zeros((config_.max_nrof_epochs, ), np.float32),
'time_validate':
np.zeros((config_.max_nrof_epochs, ), np.float32),
'time_evaluate':
np.zeros((config_.max_nrof_epochs, ), np.float32),
'prelogits_hist':
np.zeros((config_.max_nrof_epochs, 1000), np.float32),
}
for epoch in range(1, config_.max_nrof_epochs + 1):
step = sess.run(global_step, feed_dict=None)
# Train for one epoch
t = time.time()
cont = train(
config_, sess, epoch, image_list, label_list,
index_dequeue_op, enqueue_op, image_paths_placeholder,
labels_placeholder, learning_rate_placeholder,
phase_train_placeholder, batch_size_placeholder,
control_placeholder, global_step, total_loss, train_op,
summary_op, summary_writer, regularization_losses,
config_.learning_rate_schedule_file, stat,
cross_entropy_mean, accuracy, learning_rate, prelogits,
prelogits_center_loss, config_.random_rotate,
config_.random_crop, config_.random_flip, prelogits_norm,
config_.prelogits_hist_max,
config_.use_fixed_image_standardization)
stat['time_train'][epoch - 1] = time.time() - t
if not cont:
break
t = time.time()
if len(val_image_list) > 0 and (
(epoch - 1) % config_.validate_every_n_epochs
== config_.validate_every_n_epochs - 1
or epoch == config_.max_nrof_epochs):
validate(config_, sess, epoch, val_image_list,
val_label_list, enqueue_op,
image_paths_placeholder, labels_placeholder,
control_placeholder, phase_train_placeholder,
batch_size_placeholder, stat, total_loss,
regularization_losses, cross_entropy_mean,
accuracy, config_.validate_every_n_epochs,
config_.use_fixed_image_standardization)
stat['time_validate'][epoch - 1] = time.time() - t
# Save variables and the metagraph if it doesn't exist already
save_variables_and_metagraph(sess, saver, summary_writer,
model_dir, subdir, epoch)
# Evaluate on LFW
t = time.time()
if config_.lfw_dir:
evaluate(sess, enqueue_op, image_paths_placeholder,
labels_placeholder, phase_train_placeholder,
batch_size_placeholder, control_placeholder,
embeddings, label_batch, lfw_paths, actual_issame,
config_.lfw_batch_size, config_.lfw_nrof_folds,
log_dir, step, summary_writer, stat, epoch,
config_.lfw_distance_metric,
config_.lfw_subtract_mean,
config_.lfw_use_flipped_images,
config_.use_fixed_image_standardization)
stat['time_evaluate'][epoch - 1] = time.time() - t
print('Saving statistics')
return model_dir
def main(_):
if FLAGS.csv_file_path:
if os.path.exists(FLAGS.csv_file_path) is False:
csv_dir = FLAGS.csv_file_path.rsplit('/', 1)[0]
if os.path.exists(csv_dir) is False:
os.makedirs(csv_dir)
with open(FLAGS.csv_file_path, 'w') as f:
writer = csv.writer(f)
writer.writerow([
'Pruned rate', 'Acc Mean', 'Acc Std', 'Epoch No.',
'Model size through inference (MB) (Shared part + task-specific part)',
'Shared part (MB)', 'Task specific part (MB)',
'Whole masks (MB)', 'Task specific masks (MB)',
'Task specific batch norm vars (MB)',
'Task specific biases (MB)'
])
args, unparsed = parse_arguments(sys.argv[1:])
FLAGS.filters_expand_ratio = math.sqrt(FLAGS.model_size_expand_ratio)
FLAGS.history_filters_expand_ratios = [
math.sqrt(ratio) for ratio in FLAGS.history_model_size_expand_ratio
]
with tf.Graph().as_default():
with tf.Session() as sess:
# Read the file containing the pairs used for testing
pairs = lfw.read_pairs(os.path.expanduser(args.lfw_pairs))
# Get the paths for the corresponding images
paths, actual_issame = lfw.get_paths(
os.path.expanduser(args.lfw_dir), pairs)
# img = Image.open(paths[0])
image_paths_placeholder = tf.placeholder(tf.string,
shape=(None, 1),
name='image_paths')
labels_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='labels')
batch_size_placeholder = tf.placeholder(tf.int32,
name='batch_size')
control_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='control')
phase_train_placeholder = tf.placeholder(tf.bool,
name='phase_train')
nrof_preprocess_threads = 4
image_size = (args.image_size, args.image_size)
eval_input_queue = data_flow_ops.FIFOQueue(
capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1, ), (1, ), (1, )],
shared_name=None,
name=None)
eval_enqueue_op = eval_input_queue.enqueue_many(
[
image_paths_placeholder, labels_placeholder,
control_placeholder
],
name='eval_enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(
eval_input_queue, image_size, nrof_preprocess_threads,
batch_size_placeholder)
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
# Load the model
if os.path.isdir(args.model):
temp_record_file = os.path.join(args.model, 'temp_record.txt')
checkpoint_file = os.path.join(args.model, 'checkpoint')
if os.path.exists(temp_record_file) and os.path.exists(
checkpoint_file):
with open(temp_record_file) as json_file:
data = json.load(json_file)
max_acc = max(data, key=float)
epoch_no = data[max_acc]
ckpt_file = args.model + '/model-.ckpt-' + epoch_no
with open(checkpoint_file) as f:
context = f.read()
original_epoch = re.search("(\d)+", context).group()
context = context.replace(original_epoch, epoch_no)
with open(checkpoint_file, 'w') as f:
f.write(context)
if os.path.exists(os.path.join(args.model,
'copied')) is False:
os.makedirs(os.path.join(args.model, 'copied'))
copyfile(
temp_record_file,
os.path.join(args.model, 'copied', 'temp_record.txt'))
os.remove(temp_record_file)
elif os.path.exists(checkpoint_file):
ckpt = tf.train.get_checkpoint_state(args.model)
ckpt_file = ckpt.model_checkpoint_path
epoch_no = ckpt_file.rsplit('-', 1)[-1]
else:
print(
'No `temp_record.txt` or `checkpoint` in `{}`, you should pass args.model the file path, not the directory'
.format(args.model))
sys.exit(1)
else:
ckpt_file = args.model
epoch_no = ckpt_file.rsplit('-')[-1]
# Cannot use meta graph, because we need to dynamically decide batch normalization in regard to current task_id
# facenet.load_model(args.model, input_map=input_map)
# embeddings = tf.get_default_graph().get_tensor_by_name("embeddings:0")
prelogits, _ = network.inference(
image_batch,
1.0,
phase_train=phase_train_placeholder,
bottleneck_layer_size=args.embedding_size,
weight_decay=0.0)
embeddings = tf.nn.l2_normalize(prelogits,
1,
1e-10,
name='embeddings')
init_fn = slim.assign_from_checkpoint_fn(ckpt_file,
tf.global_variables())
init_fn(sess)
pruned_ratio = 0.0
model_size = 0.0
if FLAGS.print_mem or FLAGS.print_mask_info:
masks = tf.get_collection('masks')
if FLAGS.print_mask_info:
if masks:
num_elems_in_each_task_op = {}
num_elems_in_tasks_in_masks_op = {
} # two dimentional dictionary
num_elems_in_masks_op = []
num_remain_elems_in_masks_op = []
for task_id in range(1, FLAGS.task_id + 1):
num_elems_in_each_task_op[task_id] = tf.constant(
0, dtype=tf.int32)
num_elems_in_tasks_in_masks_op[task_id] = {}
# Define graph
for i, mask in enumerate(masks):
num_elems_in_masks_op.append(tf.size(mask))
num_remain_elems_in_curr_mask = tf.size(mask)
for task_id in range(1, FLAGS.task_id + 1):
cnt = tf_count(mask, task_id)
num_elems_in_tasks_in_masks_op[task_id][
i] = cnt
num_elems_in_each_task_op[task_id] = tf.add(
num_elems_in_each_task_op[task_id], cnt)
num_remain_elems_in_curr_mask -= cnt
num_remain_elems_in_masks_op.append(
num_remain_elems_in_curr_mask)
num_elems_in_network_op = tf.add_n(
num_elems_in_masks_op)
print('Calculate pruning status ...')
# Doing operation
num_elems_in_masks = sess.run(num_elems_in_masks_op)
num_elems_in_each_task = sess.run(
num_elems_in_each_task_op)
num_elems_in_tasks_in_masks = sess.run(
num_elems_in_tasks_in_masks_op)
num_elems_in_network = sess.run(
num_elems_in_network_op)
num_remain_elems_in_masks = sess.run(
num_remain_elems_in_masks_op)
# Print out the result
print('Showing pruning status ...')
if FLAGS.verbose:
for i, mask in enumerate(masks):
print('Layer %s: ' % mask.op.name, end='')
for task_id in range(1, FLAGS.task_id + 1):
cnt = num_elems_in_tasks_in_masks[task_id][
i]
print('task_%d -> %d/%d (%.2f%%), ' %
(task_id, cnt, num_elems_in_masks[i],
100 * cnt / num_elems_in_masks[i]),
end='')
print('remain -> {:.2f}%'.format(
100 * num_remain_elems_in_masks[i] /
num_elems_in_masks[i]))
print('Num elems in network: {}'.format(
num_elems_in_network))
num_elems_of_usued_weights = num_elems_in_network
for task_id in range(1, FLAGS.task_id + 1):
print('Num elems in task_{}: {}'.format(
task_id, num_elems_in_each_task[task_id]))
print('Ratio of task_{} to all: {}'.format(
task_id, num_elems_in_each_task[task_id] /
num_elems_in_network))
num_elems_of_usued_weights -= num_elems_in_each_task[
task_id]
print('Num usued elems in all masks: {}'.format(
num_elems_of_usued_weights))
# pruned_ratio = num_elems_of_usued_weights / num_elems_in_network
# print('Ratio of usused_elem to all: {}'.format(pruned_ratio))
# print('Pruning degree relative to task_{}: {:.3f}'.format(FLAGS.task_id, num_elems_of_usued_weights / (num_elems_of_usued_weights + num_elems_in_each_task[FLAGS.task_id])))
pruned_ratio_relative_to_all_elems = num_elems_of_usued_weights / num_elems_in_network
print('Ratio of usused_elem to all: {}'.format(
pruned_ratio_relative_to_all_elems))
pruned_ratio_relative_to_curr_task = num_elems_of_usued_weights / (
num_elems_of_usued_weights +
num_elems_in_each_task[FLAGS.task_id])
print('Pruning degree relative to task_{}: {:.3f}'.
format(FLAGS.task_id,
pruned_ratio_relative_to_curr_task))
if FLAGS.print_mem:
# Analyze param
start_time = time.time()
(MB_of_model_through_inference, MB_of_shared_variables,
MB_of_task_specific_variables, MB_of_whole_masks,
MB_of_task_specific_masks,
MB_of_task_specific_batch_norm_variables,
MB_of_task_specific_biases
) = model_analyzer.analyze_vars_for_current_task(
tf.model_variables(),
sess=sess,
task_id=FLAGS.task_id,
verbose=False)
duration = time.time() - start_time
print('duration time: {}'.format(duration))
if FLAGS.eval_once:
evaluate(
sess, eval_enqueue_op, image_paths_placeholder,
labels_placeholder, phase_train_placeholder,
batch_size_placeholder, control_placeholder, embeddings,
label_batch, paths, actual_issame, args.lfw_batch_size,
args.lfw_nrof_folds, args.distance_metric,
args.subtract_mean, args.use_flipped_images,
args.use_fixed_image_standardization, FLAGS.csv_file_path,
pruned_ratio, epoch_no, MB_of_model_through_inference,
MB_of_shared_variables, MB_of_task_specific_variables,
MB_of_whole_masks, MB_of_task_specific_masks,
MB_of_task_specific_batch_norm_variables,
MB_of_task_specific_biases)
def main(args):
with tf.Graph().as_default():
with tf.Session() as sess:
paths = get_dir(args.aligned_face_dir)
image_paths_placeholder = tf.placeholder(tf.string,
shape=(None, 1),
name='image_paths')
labels_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='labels')
batch_size_placeholder = tf.placeholder(tf.int32,
name='batch_size')
control_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='control')
phase_train_placeholder = tf.placeholder(tf.bool,
name='phase_train')
nrof_preprocess_threads = 4
image_size = (args.image_size, args.image_size)
#创建一个先入先出队列
eval_input_queue = data_flow_ops.FIFOQueue(
capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1, ), (1, ), (1, )],
shared_name=None,
name=None)
# 多值入队
eval_enqueue_op = eval_input_queue.enqueue_many(
[
image_paths_placeholder, labels_placeholder,
control_placeholder
],
name='eval_enqueue_op')
# 将 input_queue 中的 (image, label, control) 元祖 dequeue 出来，根据 control 里的内容
# 对 image 进行各种预处理，然后将处理后的 (image, label) 打包成真正输入 model 的 batch
image_batch, label_batch = facenet.create_input_pipeline(
eval_input_queue, image_size, nrof_preprocess_threads,
batch_size_placeholder)
# Load the model
input_map = {
'image_batch': image_batch,
'label_batch': label_batch,
'phase_train': phase_train_placeholder
}
facenet.load_model(args.model, input_map=input_map)
# Get output tensor
embeddings = tf.get_default_graph().get_tensor_by_name(
"embeddings:0")
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
nrof_images = len(paths)
labels_array = np.expand_dims(np.arange(0, nrof_images), 1)
image_paths_array = np.expand_dims(np.array(paths), 1)
control_array = np.zeros_like(labels_array, np.int32)
if args.use_fixed_image_standardization:
control_array += np.ones_like(
labels_array) * facenet.FIXED_STANDARDIZATION
sess.run(eval_enqueue_op, {image_paths_placeholder: image_paths_array, \
labels_placeholder: labels_array, control_placeholder: control_array})
embedding_size = int(embeddings.get_shape()[1])
nrof_batches = nrof_images // args.batch_size
last_batch = nrof_images % args.batch_size
emb_array = np.zeros((nrof_images, embedding_size))
lab_array = np.zeros((nrof_images, ))
for i in range(nrof_batches):
feed_dict = {
phase_train_placeholder: False,
batch_size_placeholder: args.batch_size
}
emb, lab = sess.run([embeddings, label_batch],
feed_dict=feed_dict)
lab_array[lab] = lab
emb_array[lab, :] = emb
if i % 10 == 9:
print('.', end='')
sys.stdout.flush()
feed_dict = {
phase_train_placeholder: False,
batch_size_placeholder: last_batch
}
emb, lab = sess.run([embeddings, label_batch], feed_dict=feed_dict)
lab_array[lab] = lab
emb_array[lab, :] = emb
print('')
embeddings = np.zeros((nrof_images, embedding_size))
embeddings = emb_array
embeddings0 = np.tile(embeddings[0], (len(embeddings), 1))
face_dists1 = facenet.distance(embeddings0,
embeddings,
distance_metric=1)
face_dists0 = facenet.distance(embeddings0,
embeddings,
distance_metric=0)
# face_dists = 1 - np.dot(embeddings, embeddings[2]) / \
# (np.linalg.norm(embeddings[2])*np.linalg.norm(embeddings, axis = 1))
# face_dists1 = np.arccos(face_dists) / math.pi
np.savetxt(face_dists0_path, face_dists0)
np.savetxt(face_dists1_path, face_dists1)
np.savetxt(embeddings_path, embeddings)
print(face_dists0)
print(face_dists1)
def main(args):
network = importlib.import_module(args.model_def)
image_size = (args.image_size, args.image_size)
subdir = datetime.strftime(datetime.now(), '%Y%m%d-%H%M%S')
log_dir = os.path.join(os.path.expanduser(args.logs_base_dir), subdir)
if not os.path.isdir(
log_dir): # Create the log directory if it doesn't exist
try:
os.makedirs(log_dir)
except OSError as exc:
print(exc)
model_dir = os.path.join(os.path.expanduser(args.models_base_dir), subdir)
if not os.path.isdir(
model_dir): # Create the model directory if it doesn't exist
try:
os.makedirs(model_dir)
except OSError as exc:
print(exc)
stat_file_name = os.path.join(log_dir, 'stat.h5')
# Write arguments to a text file
facenet.write_arguments_to_file(args, os.path.join(log_dir,
'arguments.txt'))
# Store some git revision info in a text file in the log directory
src_path, _ = os.path.split(os.path.realpath(__file__))
facenet.store_revision_info(src_path, log_dir, ' '.join(sys.argv))
np.random.seed(seed=args.seed)
random.seed(args.seed)
dataset = facenet.get_dataset(args.data_dir)
if args.filter_filename:
dataset = filter_dataset(dataset,
os.path.expanduser(args.filter_filename),
args.filter_percentile,
args.filter_min_nrof_images_per_class)
if args.validation_set_split_ratio > 0.0:
train_set, val_set = facenet.split_dataset(
dataset, args.validation_set_split_ratio,
args.min_nrof_val_images_per_class, 'SPLIT_IMAGES')
else:
train_set, val_set = dataset, []
nrof_classes = len(train_set)
print('Model directory: %s' % model_dir)
print('Log directory: %s' % log_dir)
pretrained_model = None
if args.pretrained_model:
pretrained_model = os.path.expanduser(args.pretrained_model)
print('Pre-trained model: %s' % pretrained_model)
if args.lfw_dir:
print('LFW directory: %s' % args.lfw_dir)
# Read the file containing the pairs used for testing
pairs = lfw.read_pairs(os.path.expanduser(args.lfw_pairs))
# Get the paths for the corresponding images
lfw_paths, actual_issame = lfw.get_paths(
os.path.expanduser(args.lfw_dir), pairs)
with tf.Graph().as_default():
tf.set_random_seed(args.seed)
global_step = tf.Variable(0, trainable=False)
# Get a list of image paths and their labels
image_list, label_list = facenet.get_image_paths_and_labels(train_set)
assert len(image_list) > 0, 'The training set should not be empty'
val_image_list, val_label_list = facenet.get_image_paths_and_labels(
val_set)
# Create a queue that produces indices into the image_list and label_list
labels = ops.convert_to_tensor(label_list, dtype=tf.int32)
range_size = array_ops.shape(labels)[0]
index_queue = tf.train.range_input_producer(range_size,
num_epochs=None,
shuffle=True,
seed=None,
capacity=32)
index_dequeue_op = index_queue.dequeue_many(
args.batch_size * args.epoch_size, 'index_dequeue')
learning_rate_placeholder = tf.placeholder(tf.float32,
name='learning_rate')
batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size')
phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train')
image_paths_placeholder = tf.placeholder(tf.string,
shape=(None, 1),
name='image_paths')
labels_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='labels')
control_placeholder = tf.placeholder(tf.int32,
shape=(None, 1),
name='control')
nrof_preprocess_threads = 4
input_queue = data_flow_ops.FIFOQueue(
capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1, ), (1, ), (1, )],
shared_name=None,
name=None)
enqueue_op = input_queue.enqueue_many(
[image_paths_placeholder, labels_placeholder, control_placeholder],
name='enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(
input_queue, image_size, nrof_preprocess_threads,
batch_size_placeholder)
image_batch = tf.identity(image_batch, 'image_batch')
image_batch = tf.identity(image_batch, 'input')
label_batch = tf.identity(label_batch, 'label_batch')
print('Number of classes in training set: %d' % nrof_classes)
print('Number of examples in training set: %d' % len(image_list))
print('Number of classes in validation set: %d' % len(val_set))
print('Number of examples in validation set: %d' % len(val_image_list))
print('Building training graph')
# Build the inference graph
prelogits, _ = network.inference(
image_batch,
args.keep_probability,
phase_train=phase_train_placeholder,
bottleneck_layer_size=args.embedding_size,
weight_decay=args.weight_decay)
logits = slim.fully_connected(
prelogits,
len(train_set),
activation_fn=None,
weights_initializer=slim.initializers.xavier_initializer(),
weights_regularizer=slim.l2_regularizer(args.weight_decay),
scope='Logits',
reuse=False)
embeddings = tf.nn.l2_normalize(prelogits, 1, 1e-10, name='embeddings')
# Norm for the prelogits
eps = 1e-4
prelogits_norm = tf.reduce_mean(
tf.norm(tf.abs(prelogits) + eps, ord=args.prelogits_norm_p,
axis=1))
tf.add_to_collection(tf.GraphKeys.REGULARIZATION_LOSSES,
prelogits_norm * args.prelogits_norm_loss_factor)
# Add center loss
prelogits_center_loss, _ = facenet.center_loss(prelogits, label_batch,
args.center_loss_alfa,
nrof_classes)
tf.add_to_collection(tf.GraphKeys.REGULARIZATION_LOSSES,
prelogits_center_loss * args.center_loss_factor)
learning_rate = tf.train.exponential_decay(
learning_rate_placeholder,
global_step,
args.learning_rate_decay_epochs * args.epoch_size,
args.learning_rate_decay_factor,
staircase=True)
tf.summary.scalar('learning_rate', learning_rate)
# Calculate the average cross entropy loss across the batch
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(
labels=label_batch,
logits=logits,
name='cross_entropy_per_example')
cross_entropy_mean = tf.reduce_mean(cross_entropy,
name='cross_entropy')
tf.add_to_collection('losses', cross_entropy_mean)
correct_prediction = tf.cast(
tf.equal(tf.argmax(logits, 1), tf.cast(label_batch, tf.int64)),
tf.float32)
accuracy = tf.reduce_mean(correct_prediction)
# Calculate the total losses
regularization_losses = tf.get_collection(
tf.GraphKeys.REGULARIZATION_LOSSES)
total_loss = tf.add_n([cross_entropy_mean] + regularization_losses,
name='total_loss')
# Build a Graph that trains the model with one batch of examples and updates the model parameters
train_op = facenet.train(total_loss, global_step, args.optimizer,
learning_rate, args.moving_average_decay,
tf.global_variables(), args.log_histograms)
# Create a saver
saver = tf.train.Saver(tf.trainable_variables(), max_to_keep=3)
# Build the summary operation based on the TF collection of Summaries.
summary_op = tf.summary.merge_all()
# Start running operations on the Graph.
gpu_options = tf.GPUOptions(
per_process_gpu_memory_fraction=args.gpu_memory_fraction)
config = tf.ConfigProto(gpu_options=gpu_options,
log_device_placement=False)
config.gpu_options.visible_device_list = str(hvd.local_rank())
sess = tf.Session(config=config)
sess.run(tf.global_variables_initializer())
sess.run(tf.local_variables_initializer())
# Horovod: broadcast initial variable states from rank 0 to all other processes.
# This is necessary to ensure consistent initialization of all workers when
# training is started with random weights or restored from a checkpoint.
bcast = hvd.broadcast_global_variables(0)
summary_writer = tf.summary.FileWriter(log_dir, sess.graph)
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
with sess.as_default():
if pretrained_model:
print('Restoring pretrained model: %s' % pretrained_model)
saver.restore(sess, pretrained_model)
# Training and validation loop
print('Running training')
nrof_steps = args.max_nrof_epochs * args.epoch_size
nrof_val_samples = int(
math.ceil(args.max_nrof_epochs / args.validate_every_n_epochs)
) # Validate every validate_every_n_epochs as well as in the last epoch
stat = {
'loss':
np.zeros((nrof_steps, ), np.float32),
'center_loss':
np.zeros((nrof_steps, ), np.float32),
'reg_loss':
np.zeros((nrof_steps, ), np.float32),
'xent_loss':
np.zeros((nrof_steps, ), np.float32),
'prelogits_norm':
np.zeros((nrof_steps, ), np.float32),
'accuracy':
np.zeros((nrof_steps, ), np.float32),
'val_loss':
np.zeros((nrof_val_samples, ), np.float32),
'val_xent_loss':
np.zeros((nrof_val_samples, ), np.float32),
'val_accuracy':
np.zeros((nrof_val_samples, ), np.float32),
'lfw_accuracy':
np.zeros((args.max_nrof_epochs, ), np.float32),
'lfw_valrate':
np.zeros((args.max_nrof_epochs, ), np.float32),
'learning_rate':
np.zeros((args.max_nrof_epochs, ), np.float32),
'time_train':
np.zeros((args.max_nrof_epochs, ), np.float32),
'time_validate':
np.zeros((args.max_nrof_epochs, ), np.float32),
'time_evaluate':
np.zeros((args.max_nrof_epochs, ), np.float32),
'prelogits_hist':
np.zeros((args.max_nrof_epochs, 1000), np.float32),
}
for epoch in range(1, args.max_nrof_epochs + 1):
step = sess.run(global_step, feed_dict=None)
# Train for one epoch
t = time.time()
cont = train(
args, sess, epoch, image_list, label_list,
index_dequeue_op, enqueue_op, image_paths_placeholder,
labels_placeholder, learning_rate_placeholder,
phase_train_placeholder, batch_size_placeholder,
control_placeholder, global_step, total_loss, train_op,
summary_op, summary_writer, regularization_losses,
args.learning_rate_schedule_file, stat, cross_entropy_mean,
accuracy, learning_rate, prelogits, prelogits_center_loss,
args.random_rotate, args.random_crop, args.random_flip,
prelogits_norm, args.prelogits_hist_max,
args.use_fixed_image_standardization)
stat['time_train'][epoch - 1] = time.time() - t
if not cont:
break
t = time.time()
if len(val_image_list) > 0 and (
(epoch - 1) % args.validate_every_n_epochs
== args.validate_every_n_epochs - 1
or epoch == args.max_nrof_epochs):
validate(args, sess, epoch, val_image_list, val_label_list,
enqueue_op, image_paths_placeholder,
labels_placeholder, control_placeholder,
phase_train_placeholder, batch_size_placeholder,
stat, total_loss, regularization_losses,
cross_entropy_mean, accuracy,
args.validate_every_n_epochs,
args.use_fixed_image_standardization)
stat['time_validate'][epoch - 1] = time.time() - t
# Save variables and the metagraph if it doesn't exist already
save_variables_and_metagraph(sess, saver, summary_writer,
model_dir, subdir, epoch)
# Evaluate on LFW
t = time.time()
if args.lfw_dir:
evaluate(sess, enqueue_op, image_paths_placeholder,
labels_placeholder, phase_train_placeholder,
batch_size_placeholder, control_placeholder,
embeddings, label_batch, lfw_paths, actual_issame,
args.lfw_batch_size, args.lfw_nrof_folds, log_dir,
step, summary_writer, stat, epoch,
args.lfw_distance_metric, args.lfw_subtract_mean,
args.lfw_use_flipped_images,
args.use_fixed_image_standardization)
stat['time_evaluate'][epoch - 1] = time.time() - t
print('Saving statistics')
with h5py.File(stat_file_name, 'w') as f:
for key, value in stat.items():
f.create_dataset(key, data=value)
return model_dir
def generate_data(feature_size, max_length, batch_size, MAT_folder, img_folder):
noise_scale = 0.15 # 0.15
# connect_thresh = 0.95
connect_thresh = np.random.uniform(0.6, 1)
# sample_p = np.array([0.0852,0.1996,0.2550,0.0313,0.0854,0.1546,0.1890])
sample_p = np.array(sample_prob)
sample_p = list(sample_p / sum(sample_p))
# load mat files
Mat_paths = sorted(os.listdir(MAT_folder))
choose_idx = np.random.choice(len(Mat_paths), size=batch_size, p=sample_p)
Mat_files = []
for n in range(batch_size):
temp_path = MAT_folder + '/' + Mat_paths[choose_idx[n]]
# temp_mat_file = loadmat(temp_path)
f = open(temp_path, 'rb')
gt_info = pickle.load(f)
Mat_files.append({'gtInfo': gt_info})
# X = np.zeros((batch_size,1,max_length,1+4+512))
X = np.zeros((batch_size, feature_size, max_length, 3))
Y = np.zeros((batch_size, 2))
crop_bbox = []
# positive
for n in range(int(batch_size / 2)):
print('positive', n, Mat_paths[choose_idx[n]], Mat_files[n]['gtInfo']['X'].shape)
fr_num = Mat_files[n]['gtInfo']['X'].shape[0]
id_num = Mat_files[n]['gtInfo']['X'].shape[1]
Y[n, 0] = 1
X_2d = Mat_files[n]['gtInfo']['X']
Y_2d = Mat_files[n]['gtInfo']['Y']
W_2d = Mat_files[n]['gtInfo']['W']
H_2d = Mat_files[n]['gtInfo']['H']
#########################################
X_2d = X_2d - margin * W_2d
Y_2d = Y_2d - margin * H_2d
W_2d = (1 + 2 * margin) * W_2d
H_2d = (1 + 2 * margin) * H_2d
##########################################
if len(Mat_files[n]['gtInfo']) <= 4:
V_2d = []
else:
V_2d = Mat_files[n]['gtInfo']['V']
if len(Mat_files[n]['gtInfo']) == 6:
img_size = Mat_files[n]['gtInfo']['img_size']
else:
img_size = [1920, 1080]
while 1:
obj_id = np.random.randint(id_num, size=1)[0]
bbox_tracklet, t_interval, err_flag = split_track(X_2d, Y_2d, W_2d, H_2d, V_2d, img_size, obj_id,
noise_scale, connect_thresh)
if err_flag == 1:
continue
cand_pairs = []
if len(bbox_tracklet) <= 1:
continue
for k1 in range(len(bbox_tracklet) - 1):
for k2 in range(k1 + 1, len(bbox_tracklet)):
if t_interval[k1, 0] + max_length > t_interval[k2, 0]:
t_dist = t_interval[k2, 0] - t_interval[k1, 1]
cand_pairs.append([k1, k2, t_dist])
if len(cand_pairs) == 0:
continue
cand_pairs = np.array(cand_pairs)
rand_num = np.random.rand(1)[0]
# print(rand_num)
if rand_num < 0.7:
select_p = np.exp(-np.power(cand_pairs[:, 2], 2) / 100)
select_p = select_p / sum(select_p)
# print(select_p)
pair_idx = np.random.choice(len(cand_pairs), size=1, p=select_p)[0]
else:
pair_idx = np.random.randint(len(cand_pairs), size=1)[0]
select_pair = cand_pairs[pair_idx]
select_pair = select_pair.astype(int)
abs_fr_t1 = int(t_interval[select_pair[0], 0])
abs_fr_t2 = int(t_interval[select_pair[0], 1])
abs_fr_t3 = int(t_interval[select_pair[1], 0])
abs_fr_t4 = int(min(abs_fr_t1 + max_length - 1, t_interval[select_pair[1], 1]))
t1 = 0
t2 = abs_fr_t2 - abs_fr_t1
t3 = abs_fr_t3 - abs_fr_t1
t4 = abs_fr_t4 - abs_fr_t1
# mask
X[n, :, t1:t2 + 1, 1] = 1
X[n, :, t3:t4 + 1, 2] = 1
# X[n,4:,t1:t2+1,1] = bbox_tracklet[select_pair[0]][:,5]
# X[n,4:,t3:t4+1,2] = bbox_tracklet[select_pair[1]][0:t4-t3+1,5]
# X
X[n, 0, t1:t2 + 1, 0] = 0.5 * (bbox_tracklet[select_pair[0]][:, 1] + bbox_tracklet[select_pair[0]][:, 3]) / \
img_size[0]
X[n, 0, t3:t4 + 1, 0] = 0.5 * (
bbox_tracklet[select_pair[1]][0:t4 - t3 + 1, 1] + bbox_tracklet[select_pair[1]][0:t4 - t3 + 1,
3]) / img_size[0]
# Y
X[n, 1, t1:t2 + 1, 0] = 0.5 * (bbox_tracklet[select_pair[0]][:, 2] + bbox_tracklet[select_pair[0]][:, 4]) / \
img_size[1]
X[n, 1, t3:t4 + 1, 0] = 0.5 * (
bbox_tracklet[select_pair[1]][0:t4 - t3 + 1, 2] + bbox_tracklet[select_pair[1]][0:t4 - t3 + 1,
4]) / img_size[1]
# W
X[n, 2, t1:t2 + 1, 0] = (bbox_tracklet[select_pair[0]][:, 3] - bbox_tracklet[select_pair[0]][:, 1]) / \
img_size[0]
X[n, 2, t3:t4 + 1, 0] = (bbox_tracklet[select_pair[1]][0:t4 - t3 + 1, 3] - bbox_tracklet[select_pair[1]][
0:t4 - t3 + 1, 1]) / img_size[0]
# H
X[n, 3, t1:t2 + 1, 0] = (bbox_tracklet[select_pair[0]][:, 4] - bbox_tracklet[select_pair[0]][:, 2]) / \
img_size[1]
X[n, 3, t3:t4 + 1, 0] = (bbox_tracklet[select_pair[1]][0:t4 - t3 + 1, 4] - bbox_tracklet[select_pair[1]][
0:t4 - t3 + 1, 2]) / img_size[1]
'''
plt.plot(X[n,0,:,0], 'ro')
plt.show()
plt.plot(X[n,1,:,0], 'ro')
plt.show()
plt.plot(X[n,2,:,0], 'ro')
plt.show()
plt.plot(X[n,3,:,0], 'ro')
plt.show()
plt.plot(X[n,0,:,1], 'ro')
plt.show()
plt.plot(X[n,0,:,2], 'ro')
plt.show()
'''
# save all bbox
temp_crop_bbox = np.concatenate(
(bbox_tracklet[select_pair[0]], bbox_tracklet[select_pair[1]][0:t4 - t3 + 1, :]), axis=0)
temp_crop_bbox = temp_crop_bbox.astype(int)
crop_bbox.append(temp_crop_bbox)
break
# negative
for n in range(int(batch_size / 2), batch_size):
print('negative', n, Mat_paths[choose_idx[n]], Mat_files[n]['gtInfo']['X'].shape)
fr_num = Mat_files[n]['gtInfo']['X'].shape[0]
id_num = Mat_files[n]['gtInfo']['X'].shape[1]
Y[n, 1] = 1
X_2d = Mat_files[n]['gtInfo']['X']
Y_2d = Mat_files[n]['gtInfo']['Y']
W_2d = Mat_files[n]['gtInfo']['W']
H_2d = Mat_files[n]['gtInfo']['H']
#########################################
X_2d = X_2d - margin * W_2d
Y_2d = Y_2d - margin * H_2d
W_2d = (1 + 2 * margin) * W_2d
H_2d = (1 + 2 * margin) * H_2d
##########################################
if len(Mat_files[n]['gtInfo']) <= 4:
V_2d = []
else:
V_2d = Mat_files[n]['gtInfo']['V']
if len(Mat_files[n]['gtInfo']) == 6:
img_size = Mat_files[n]['gtInfo']['img_size']
else:
img_size = [1920, 1080]
# check candidate obj pairs
# pair_mat = np.zeros((id_num,id_num))
cand_idx_pairs = []
for n1 in range(id_num - 1):
for n2 in range(n1 + 1, id_num):
cand_fr1 = np.where(W_2d[:, n1] > 0)[0]
cand_fr2 = np.where(W_2d[:, n2] > 0)[0]
if max(cand_fr1[0], cand_fr2[0]) < min(cand_fr1[-1], cand_fr2[-1]): # cand_fr1 and cand_fr2 are overlapped
cand_idx_pairs.append([n1, n2])
# pair_mat[n1,n2] = 1
# cand_pairs = np.nonzero(pair_mat)
while 1:
if len(cand_idx_pairs) == 0:
import pdb
pdb.set_trace()
pair_idx = np.random.randint(len(cand_idx_pairs), size=1)[0]
obj_id1 = cand_idx_pairs[pair_idx][0]
obj_id2 = cand_idx_pairs[pair_idx][1]
part_W_mat1 = W_2d[:, obj_id1]
non_zero_idx1 = np.where(part_W_mat1 > 0)[0]
part_W_mat2 = W_2d[:, obj_id2]
non_zero_idx2 = np.where(part_W_mat2 > 0)[0]
if len(non_zero_idx1) == 0 or len(non_zero_idx2) == 0 or \
max(non_zero_idx1) + max_length < min(non_zero_idx2) or min(non_zero_idx1) > max(non_zero_idx2):
continue
bbox_tracklet1, t_interval1, err_flag = split_track(X_2d, Y_2d, W_2d, H_2d, V_2d, img_size, obj_id1,
noise_scale, connect_thresh)
if err_flag == 1:
continue
bbox_tracklet2, t_interval2, err_flag = split_track(X_2d, Y_2d, W_2d, H_2d, V_2d, img_size, obj_id2,
noise_scale, connect_thresh)
if err_flag == 1:
continue
cand_pairs = []
if len(bbox_tracklet1) <= 1 or len(bbox_tracklet2) <= 1:
continue
for k1 in range(len(bbox_tracklet1)):
for k2 in range(len(bbox_tracklet2)):
if t_interval1[k1, 0] + max_length > t_interval2[k2, 0] and t_interval1[k1, 1] < t_interval2[k2, 0]:
t_dist = t_interval2[k2, 0] - t_interval1[k1, 1]
cand_pairs.append([k1, k2, t_dist])
if len(cand_pairs) == 0:
continue
cand_pairs = np.array(cand_pairs)
rand_num = np.random.rand(1)[0]
# print(rand_num)
if rand_num < 0.7:
select_p = np.exp(-np.power(cand_pairs[:, 2], 2) / 100)
select_p = select_p / sum(select_p)
# print(select_p)
pair_idx = np.random.choice(len(cand_pairs), size=1, p=select_p)[0]
else:
pair_idx = np.random.randint(len(cand_pairs), size=1)[0]
select_pair = cand_pairs[pair_idx]
select_pair = select_pair.astype(int)
abs_fr_t1 = int(t_interval1[select_pair[0], 0])
abs_fr_t2 = int(t_interval1[select_pair[0], 1])
abs_fr_t3 = int(t_interval2[select_pair[1], 0])
abs_fr_t4 = int(min(abs_fr_t1 + max_length - 1, t_interval2[select_pair[1], 1]))
t1 = 0
t2 = abs_fr_t2 - abs_fr_t1
t3 = abs_fr_t3 - abs_fr_t1
t4 = abs_fr_t4 - abs_fr_t1
# mask
X[n, :, t1:t2 + 1, 1] = 1
X[n, :, t3:t4 + 1, 2] = 1
# X[n,4:,t1:t2+1,1] = bbox_tracklet1[select_pair[0]][:,5]
# X[n,4:,t3:t4+1,2] = bbox_tracklet2[select_pair[1]][0:t4-t3+1,5]
# X
X[n, 0, t1:t2 + 1, 0] = 0.5 * (
bbox_tracklet1[select_pair[0]][:, 1] + bbox_tracklet1[select_pair[0]][:, 3]) / img_size[0]
X[n, 0, t3:t4 + 1, 0] = 0.5 * (
bbox_tracklet2[select_pair[1]][0:t4 - t3 + 1, 1] + bbox_tracklet2[select_pair[1]][0:t4 - t3 + 1,
3]) / img_size[0]
# Y
X[n, 1, t1:t2 + 1, 0] = 0.5 * (
bbox_tracklet1[select_pair[0]][:, 2] + bbox_tracklet1[select_pair[0]][:, 4]) / img_size[1]
X[n, 1, t3:t4 + 1, 0] = 0.5 * (
bbox_tracklet2[select_pair[1]][0:t4 - t3 + 1, 2] + bbox_tracklet2[select_pair[1]][0:t4 - t3 + 1,
4]) / img_size[1]
# W
X[n, 2, t1:t2 + 1, 0] = (bbox_tracklet1[select_pair[0]][:, 3] - bbox_tracklet1[select_pair[0]][:, 1]) / \
img_size[0]
X[n, 2, t3:t4 + 1, 0] = (bbox_tracklet2[select_pair[1]][0:t4 - t3 + 1, 3] - bbox_tracklet2[select_pair[1]][
0:t4 - t3 + 1, 1]) / img_size[0]
# H
X[n, 3, t1:t2 + 1, 0] = (bbox_tracklet1[select_pair[0]][:, 4] - bbox_tracklet1[select_pair[0]][:, 2]) / \
img_size[1]
X[n, 3, t3:t4 + 1, 0] = (bbox_tracklet2[select_pair[1]][0:t4 - t3 + 1, 4] - bbox_tracklet2[select_pair[1]][
0:t4 - t3 + 1, 2]) / img_size[1]
'''
plt.plot(X[n,0,:,0], 'ro')
plt.show()
plt.plot(X[n,1,:,0], 'ro')
plt.show()
plt.plot(X[n,2,:,0], 'ro')
plt.show()
plt.plot(X[n,3,:,0], 'ro')
plt.show()
plt.plot(X[n,0,:,1], 'ro')
plt.show()
plt.plot(X[n,0,:,2], 'ro')
plt.show()
'''
# save all bbox
temp_crop_bbox = np.concatenate(
(bbox_tracklet1[select_pair[0]], bbox_tracklet2[select_pair[1]][0:t4 - t3 + 1, :]), axis=0)
temp_crop_bbox = temp_crop_bbox.astype(int)
crop_bbox.append(temp_crop_bbox)
break
# crop data to a temp folder
print('cropping data to tmp folder ...')
if not os.path.exists(temp_folder):
os.makedirs(temp_folder)
all_paths = []
for n in range(batch_size):
temp_all_path = []
seq_name = Mat_paths[choose_idx[n]][:-4]
# img_path = img_folder + '/' + seq_name + '/img1/'
img_path = img_folder + '/' + seq_name + '/'
track_name = file_name(n + 1, 4)
save_path = temp_folder + '/' + track_name
if not os.path.exists(save_path):
os.makedirs(save_path)
for k in range(len(crop_bbox[n])):
fr_id = crop_bbox[n][k, 0] + 1
temp_img_path = img_path + file_name(fr_id, 6) + '.jpg'
img = scipy.ndimage.imread(temp_img_path)
bbox_img = img[crop_bbox[n][k, 2]:crop_bbox[n][k, 4], crop_bbox[n][k, 1]:crop_bbox[n][k, 3], :]
bbox_img = scipy.misc.imresize(bbox_img, size=(bbox_size, bbox_size))
bbox_img_path = save_path + '/' + file_name(k, 4) + '.png'
temp_all_path.append(bbox_img_path)
scipy.misc.imsave(bbox_img_path, bbox_img)
all_paths.append(temp_all_path)
print('start training ...')
f_image_size = 160
distance_metric = 0
with tf.Graph().as_default():
with tf.Session() as sess:
image_paths_placeholder = tf.placeholder(tf.string, shape=(None, 1), name='image_paths')
labels_placeholder = tf.placeholder(tf.int32, shape=(None, 1), name='labels')
batch_size_placeholder = tf.placeholder(tf.int32, name='batch_size')
control_placeholder = tf.placeholder(tf.int32, shape=(None, 1), name='control')
phase_train_placeholder = tf.placeholder(tf.bool, name='phase_train')
nrof_preprocess_threads = 4
image_size = (f_image_size, f_image_size)
eval_input_queue = data_flow_ops.FIFOQueue(capacity=2000000,
dtypes=[tf.string, tf.int32, tf.int32],
shapes=[(1,), (1,), (1,)],
shared_name=None, name=None)
eval_enqueue_op = eval_input_queue.enqueue_many(
[image_paths_placeholder, labels_placeholder, control_placeholder], name='eval_enqueue_op')
image_batch, label_batch = facenet.create_input_pipeline(eval_input_queue, image_size,
nrof_preprocess_threads, batch_size_placeholder)
# Load the model
input_map = {'image_batch': image_batch, 'label_batch': label_batch, 'phase_train': phase_train_placeholder}
facenet.load_model(triplet_model, input_map=input_map)
# Get output tensor
embeddings = tf.get_default_graph().get_tensor_by_name("embeddings:0")
#
coord = tf.train.Coordinator()
tf.train.start_queue_runners(coord=coord, sess=sess)
for n in range(len(all_paths)):
# print(n)
lfw_batch_size = len(all_paths[n])
emb_array = evaluate(sess, eval_enqueue_op, image_paths_placeholder, labels_placeholder,
phase_train_placeholder,
batch_size_placeholder, control_placeholder, embeddings, label_batch, all_paths[n],
lfw_batch_size, distance_metric)
if X[n, 4:, X[n, 0, :, 1] + X[n, 0, :, 2] > 0.5, 0].shape[0] != emb_array.shape[0]:
aa = 0
import pdb
pdb.set_trace()
X[n, 4:, X[n, 0, :, 1] + X[n, 0, :, 2] > 0.5, 0] = emb_array
return X, Y