) ##這個函數是用來讀TFRecord文件的，dataset中的每一個元素就是一個TFExample，裡面放filename，是string A Dataset comprising records from one or more TFRecord files.
dataset = dataset.map(
parse_function
) ##parse the whole batch at once. Apply this function to each item in the dataset using the tf.data.Dataset.map method:處理 string，將 string 轉化為 tf.Tensor 物件
dataset = dataset.shuffle(buffer_size=args.buffer_size) ##打亂
dataset = dataset.batch(args.batch_size) ##設定batch
iterator = dataset.make_initializable_iterator(
) ##構建 iterator。Initializable iterator 要求在使用之前顯式的通過呼叫iterator.initializer操作初始化，這使得在定義資料集時可以結合tf.placeholder傳入引數
next_element = iterator.get_next()
# 2.2 prepare validate datasets
ver_list = []
ver_name_list = []
for db in args.eval_datasets:
print('begin db %s convert.' % db) #列出每一個驗證dataset的path，一次迴圈一個
data_set = load_bin(
db, args.image_size,
args) #預設是lfw,image size 27行 112*112，得到dataset BGR [112, 112, 3]
ver_list.append(data_set) #把驗證集加到ver_list裡
ver_name_list.append(db) #把驗證集path加到ver_name_list裡
# 3. define network, loss, optimize method, learning rate schedule, summary writer, saver
# 3.1 inference phase
w_init_method = tf.contrib.layers.xavier_initializer(uniform=False)
net = get_resnet(images,
args.net_depth,
type='ir',
w_init=w_init_method,
trainable=True,
keep_rate=dropout_rate)
# 3.2 get arcface loss
logit = arcface_loss(embedding=net.outputs,
labels=labels,
tfrecords_f = os.path.join(args.tfrecords_file_path, 'tran.tfrecords')
dataset = tf.data.TFRecordDataset(tfrecords_f)
dataset = dataset.map(parse_function).shuffle(
buffer_size=args.buffer_size).batch(args.batch_size).prefetch(
args.batch_size)
# dataset = dataset.map(parse_function)
# dataset = dataset.shuffle(buffer_size=args.buffer_size)
# dataset = dataset.batch(args.batch_size)
iterator = dataset.make_initializable_iterator()
next_element = iterator.get_next()
# 2.2 prepare validate datasets
ver_list = []
ver_name_list = []
for db in args.eval_datasets:
print('begin db %s convert.' % db)
data_set = load_bin(db, args.image_size, args)
ver_list.append(data_set)
ver_name_list.append(db)
# 3. define network, loss, optimize method, learning rate schedule, summary writer, saver
# 3.1 inference phase
w_init_method = tf.contrib.layers.xavier_initializer(uniform=False)
net = get_resnet(images,
args.net_depth,
type='ir',
w_init=w_init_method,
trainable=True,
keep_rate=dropout_rate)
# 3.2 get arcface loss
logit = arcface_loss(embedding=net.outputs,
labels=labels,
w_init=w_init_method,
# 2.1 train datasets
# the image is substracted 127.5 and multiplied 1/128.
# random flip left right
tfrecords_f = os.path.join(args.tfrecords_file_path, 'tran.tfrecords')
dataset = tf.data.TFRecordDataset(tfrecords_f)
dataset = dataset.map(parse_function)
dataset = dataset.shuffle(buffer_size=args.buffer_size)
dataset = dataset.batch(args.batch_size)
iterator = dataset.make_initializable_iterator()
next_element = iterator.get_next()
# 2.2 prepare validate datasets
ver_list = []
ver_name_list = []
for db in args.eval_datasets:
print('begin db %s convert.' % db)
data_set = load_bin(db, args.image_size, args)
ver_list.append(data_set)
ver_name_list.append(db)
# 3. define network, loss, optimize method, learning rate schedule, summary writer, saver
# 3.1 inference phase
w_init_method = tf.contrib.layers.xavier_initializer(uniform=False)
# 3.2 define the learning rate schedule
p = int(512.0/args.batch_size)
lr_steps = [p*val for val in args.lr_steps]
print('learning rate steps: ', lr_steps)
lr = tf.train.piecewise_constant(global_step, boundaries=lr_steps, values=[0.001, 0.0005, 0.0003, 0.0001],
name='lr_schedule')
# 3.3 define the optimize method
opt = tf.train.MomentumOptimizer(learning_rate=lr, momentum=args.momentum)
# Calculate the gradients for each model tower.
) # map，parse_function函数对每一个图进行处理，bgr位置转换，标准化，随机数据增强
else:
dataset = dataset.map(folder_parse_function)
dataset = dataset.shuffle(buffer_size=buffer_size) # shuffle
dataset = dataset.batch(batch_size) # ((?, 112, 112, 3)， (?,))
iterator = dataset.make_initializable_iterator(
) # ((?, 112, 112, 3)， (?,))
next_element = iterator.get_next(
) # <class 'tuple'>: (<tf.Tensor 'IteratorGetNext:0' shape=(?, 112, 112, 3) dtype=float32>, <tf.Tensor 'IteratorGetNext:1' shape=(?,) dtype=int64>)
# 2.2 prepare validate datasets
ver_list = []
ver_name_list = []
for db in eval_datasets:
print('begin db %s convert.' % db)
data_set = load_bin(
db, image_size, eval_db_path
) # [正(12000, 112, 112, 3), 反(12000, 112, 112, 3)], 6000
ver_list.append(data_set)
ver_name_list.append(db)
# 3. define network, loss, optimize method, learning rate schedule, summary writer, saver
# 3.1 inference phase
w_init_method = tf.contrib.layers.xavier_initializer(uniform=False)
net = get_resnet(images,
net_depth,
type='ir',
w_init=w_init_method,
trainable=True,
keep_rate=dropout_rate)
# 3.2 get arcface loss
logit = arcface_loss(embedding=net.outputs,
labels=labels,
parser.add_argument('--ckpt_index_list',
default=['1950000.ckpt'],
help='ckpt file indexes')
args = parser.parse_args()
return args
if __name__ == '__main__':
args = get_args()
ver_list = []
ver_name_list = []
for db in args.eval_datasets:
print('begin db %s convert.' % db)
data_set = load_bin(
db, args.image_size, args.eval_db_path
) # (data_list, issame_list)，len(data_list)=2 ,data_list[0].shape=(12000, 112, 112, 3), len(issame_list) = 6000
ver_list.append(data_set)
ver_name_list.append(db)
images = tf.placeholder(name='img_inputs',
shape=[None, *args.image_size, 3],
dtype=tf.float32) # (?, 112, 112, 3)
labels = tf.placeholder(name='img_labels', shape=[
None,
], dtype=tf.int64) # (?, )
dropout_rate = tf.placeholder(name='dropout_rate', dtype=tf.float32)
w_init_method = tf.contrib.layers.xavier_initializer(
uniform=False) # 随机初始化权重先把空架子搭起来，后续再往里面restore train好的权重
net = get_resnet(
]:
dataset = dataset.map(
raw_parse_function
) # map，parse_function函数对每一个图进行处理，bgr位置转换，标准化，随机数据增强
else:
dataset = dataset.map(folder_parse_function)
dataset = dataset.shuffle(buffer_size=buffer_size)
dataset = dataset.batch(batch_size)
iterator = dataset.make_initializable_iterator()
next_element = iterator.get_next()
# 2.2 prepare validate datasets
ver_list = []
ver_name_list = []
for db in eval_datasets:
print('begin db %s convert.' % db)
data_set = load_bin(db, image_size, eval_db_path)
ver_list.append(data_set)
ver_name_list.append(db)
# 3. define network, loss, optimize method, learning rate schedule, summary writer, saver
# 3.1 inference phase
w_init_method = tf.contrib.layers.xavier_initializer(uniform=False)
# 3.2 define the learning rate schedule
p = int(512.0 / batch_size)
lr_steps = [p * val for val in lr_steps]
print('learning rate steps: ', lr_steps)
lr = tf.train.piecewise_constant(global_step,
boundaries=lr_steps,
values=lr_values,
name='lr_schedule')
# 3.3 define the optimize method
opt = tf.train.MomentumOptimizer(learning_rate=lr, momentum=momentum)
