def run_train():
train_dir='C:\\Users\\fengc\\Desktop\\The_simpsons_train_data'
logs_train_dir='C:\\Users\\fengc\\Desktop\\simpsons_3rd_version\\record\\May17_1'
train,train_label=image_input.get_files(train_dir)
train_batch,train_label_batch=image_input.get_batch(train,
train_label,
img_W,
img_H,
batch_size,
capacity)
train_logits=image_model1.inference(train_batch,batch_size,n_classes)
train_loss=image_model1.losses(train_logits,train_label_batch)
train_op=image_model1.trainning(train_loss,learning_rate)
train_acc=image_model1.evaluation(train_logits,train_label_batch)
summary_op=tf.summary.merge_all()
sess=tf.Session()
train_writer=tf.summary.FileWriter(logs_train_dir, sess.graph)
init=tf.global_variables_initializer()
sess.run(init)
saver=tf.train.Saver()
coord=tf.train.Coordinator()
threads=tf.train.start_queue_runners(sess=sess,coord=coord)
try:
for step in range(max_step):
if coord.should_stop():
break
#_,tra_loss,tra_acc=sess.run([train_op,train_loss,train_acc])
[_, tra_loss, tra_acc] = sess.run([train_op, train_loss, train_acc])
#tf.variable_scope.reuse_variables()
if step % 10 ==0:
print('Step %d, train loss = %.2f, train accuracy = %.4f' %(step,tra_loss,tra_acc))
summary_str=sess.run(summary_op)
train_writer.add_summary(summary_str,step)
if step % 100 == 0 or (step+1)==max_step:
checkpoint_path=os.path.join(logs_train_dir,'model.ckpt')
saver.save(sess,checkpoint_path,global_step=step)
except tf.errors.OutOfRangeError:
print('Done training -- epoch limit reached')
finally:
coord.request_stop()
coord.join(threads)
sess.close()
def train():
# train_dir='/media/vcca/新梗结衣/extendck48_chose'
# train_dir='D:\Leung\lenet5\data\ck48_train'
# tes_dir='/media/vcca/新梗结衣/emotion/ck48_tes'
is_training = tf.placeholder(tf.bool)
train, train_label, val, val_label = image_input.getfrom_raf(file_dir,
True,
ratio=0.1)
# train, train_label, val, val_label = image_input.get_file(train_dir,True,0.1)
train_batch, train_label_batch = image_input.get_batch(
train, train_label, BATCH_SIZE, True)
val_batch, val_label_batch = image_input.get_batch(val, val_label,
BATCH_SIZE, False)
#pre_logits output nodes 1024
pre_logits = list_model.inference(x,
N_CLASSES,
is_train=is_training,
keep_prob=keep_prob)
#center_loss and center_op
center_loss, centers_update_op = list_model.center_loss(
pre_logits, y_, 0.5, 7)
logits = tool.FC_Layer('fc7', pre_logits, out_nodes=7)
loss = list_model.losses(logits, y_)
total_loss = loss + 0.05 * center_loss
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
with tf.control_dependencies([centers_update_op]):
train_op, lr_rate = list_model.training(loss, learning_rate)
acc = list_model.evaluation(logits, y_)
# with tf.Session() as sess:
with tf.Session(config=tf.ConfigProto(allow_soft_placement=True,
log_device_placement=True)) as sess:
# with tf.device("/gpu:0"):
#store variable in the batch_norm
var_list = tf.trainable_variables()
g_list = tf.global_variables()
bn_moving_vars = [g for g in g_list if 'moving_mean' in g.name]
bn_moving_vars += [g for g in g_list if 'moving_variance' in g.name]
var_list += bn_moving_vars
saver = tf.train.Saver(var_list=var_list, max_to_keep=5)
# saver = tf.train.Saver()
sess.run(tf.global_variables_initializer())
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(coord=coord)
summary_op = tf.summary.merge_all()
train_writer = tf.summary.FileWriter(log_dir, sess.graph)
# val_writer = tf.summary.FileWriter(logs_val_dir, sess.graph)
try:
for step in np.arange(MAX_STEP):
if coord.should_stop():
break
# input pipleline
# _, tra_loss,tra_acc = sess.run([train_op, loss, acc])
tra_images, tra_labels = sess.run(
[train_batch, train_label_batch])
_, _ops, tra_loss, tra_acc, summary_str, lr = sess.run(
[
train_op, update_ops, total_loss, acc, summary_op,
lr_rate
],
feed_dict={
x: tra_images,
y_: tra_labels,
is_training: True,
keep_prob: 0.4
})
if step % 10 == 0:
print(
'%s,Step %d, train loss = %.6f, train accuracy = %.5f ,lr = %f'
% (datetime.now(), step, tra_loss, tra_acc, lr))
# summary_str = sess.run(summary_op)
train_writer.add_summary(summary_str, step)
if step % 100 == 0 or (step + 1) == MAX_STEP:
# sess.run(val_batch)
# _, val_loss,val_acc = sess.run([train_op, loss, acc])
val_images, val_labels = sess.run(
[val_batch, val_label_batch])
val_loss, val_acc = sess.run(
[total_loss, acc],
feed_dict={
x: val_images,
y_: val_labels,
is_training: False,
keep_prob: 1.0
})
print(
'**%s, Step %d, val loss = %.6f, val accuracy = %.5f **'
% (datetime.now(), step, val_loss, val_acc))
# summary_str = sess.run(summary_op)
# val_writer.add_summary(summary_str, step)
if step % 1000 == 0 or (step + 1) == MAX_STEP:
checkpoint_path = os.path.join(log_dir, 'model.ckpt')
saver.save(sess, checkpoint_path, global_step=step)
except tf.errors.OutOfRangeError:
print('Done training -- epoch limit reached')
finally:
coord.request_stop()
coord.join(threads)
def svm_evaluate():
file_dir = 'D:\Leung\RAF-DB\list_patition_label.txt'
with tf.Graph().as_default():
BATCH_SIZE = 32
x = tf.placeholder(tf.float32, shape=[BATCH_SIZE, IMG_W, IMG_H, 1])
y_ = tf.placeholder(tf.int32, shape=[BATCH_SIZE])
is_training = tf.placeholder(tf.bool)
keep_prob = tf.placeholder("float")
x_train = np.zeros((1, 2048))
y_train = []
x_test = np.zeros((1, 2048))
y_test = []
# pred_label=[]
# true_label=[]
# tes_dir='D:\Leung\lenet5\data\ck48_tes'
train, train_label, val, val_label = image_input.getfrom_raf(
file_dir, True)
tes_images, tes_labels = image_input.getfrom_raf(file_dir, False)
# tes_images,tes_labels = image_input.get_file(tes_dir,False,0)
n_test = len(tes_labels)
n_train = len(train_label)
tes_batch, label_batch = image_input.get_batch(tes_images, tes_labels,
BATCH_SIZE, False)
tra_batch, tra_label_batch = image_input.get_batch(
train, train_label, BATCH_SIZE, False)
pre_logits = list_model.inference(x,
N_CLASSES,
is_train=is_training,
keep_prob=keep_prob)
saver = tf.train.Saver(tf.global_variables())
with tf.device("/gpu:0"):
with tf.Session() as sess:
print("Reading checkpoints...")
ckpt = tf.train.get_checkpoint_state(log_dir)
if ckpt and ckpt.model_checkpoint_path:
# ckpt.model_checkpoint_path='D:\Leung\lenet5\log\log0928_alone_conv4\model.ckpt-11999'
global_step = ckpt.model_checkpoint_path.split(
'\\')[-1].split('-')[-1]
saver.restore(sess, ckpt.model_checkpoint_path)
print('Loading success, global_step is %s' % global_step)
else:
print('No checkpoint file found')
return
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
try:
tra_num_iter = int(math.ceil(n_train / BATCH_SIZE))
tes_num_iter = int(math.ceil(n_test / BATCH_SIZE))
# true_count = 0
# total_sample_count = num_iter * BATCH_SIZE
step = 0
t0 = time()
# while step<1:
print("Start extract tra_features!")
while step < tra_num_iter and not coord.should_stop():
tra_image, tra_label = sess.run(
[tra_batch, tra_label_batch])
feature = sess.run(
[pre_logits],
feed_dict={
x: tra_image,
y_: tra_label,
is_training: False,
keep_prob: 0.4
})
feature = np.array(feature)
feature = feature.reshape((32, 2048))
# print('step: %s' %step)
# print(feature.shape)
step += 1
# feature = np.reshape(feature,[1,2048])
x_train = np.vstack((x_train, feature))
y_train = np.append(y_train, tra_label)
# x_train=np.reshape()
print("done in %0.3fs" % (time() - t0))
print(x_train.shape, y_train.shape)
print("Start extract tes_features!")
t0 = time()
step = 0
while step < tes_num_iter and not coord.should_stop():
tes_image, tes_label = sess.run(
[tes_batch, label_batch])
feature = sess.run(
[pre_logits],
feed_dict={
x: tes_image,
y_: tes_label,
is_training: False,
keep_prob: 1.0
})
feature = np.array(feature)
feature = feature.reshape((32, 2048))
# print('step: %s' %step)
# print(feature.shape)
x_test = np.vstack((x_test, feature))
y_test = np.append(y_test, tes_label)
step += 1
print("done in %0.3fs" % (time() - t0))
print(x_test.shape, y_test.shape)
# list_model.plot_confusion_matrix(pred_label,true_label)
except Exception as e:
coord.request_stop(e)
finally:
coord.request_stop()
coord.join(threads)
print("Fitting the classifier to the training set")
t0 = time()
# param_grid = {'C': [1e3, 1e5],
# 'gamma': [0.005, 0.01], }
clf = SVC(C=1.0,
kernel='rbf',
gamma=0.012,
cache_size=2000,
class_weight='balanced',
decision_function_shape='ovr')
# clf = GridSearchCV(SVC(kernel='rbf',cache_size=2000, class_weight='balanced'), param_grid)
clf = clf.fit(x_train[1:-17, ], y_train[:-17])
print("fit classfier done in %0.3fs" % (time() - t0))
print(clf)
# print("Best estimator found by grid search:")
# print(clf.best_estimator_)
t0 = time()
y_pred = clf.predict(x_test[1:-4, ])
print("done in %0.3fs" % (time() - t0))
list_model.plot_confusion_matrix(y_pred, y_test[:-4])
def evaluate():
# file_dir='D:\Leung\lenet5\data\list_patition_label.txt'
with tf.Graph().as_default():
# x = tf.placeholder(tf.float32, shape=[BATCH_SIZE, IMG_W, IMG_H, 1])
# y_ = tf.placeholder(tf.int32, shape=[BATCH_SIZE])
is_training = tf.placeholder(tf.bool)
keep_prob = tf.placeholder("float")
count = 0
pred_label = []
true_label = []
# tes_dir='D:\Leung\lenet5\data\ck48_tes'
tes_images, tes_labels = image_input.getfrom_raf(file_dir, False)
# tes_images,tes_labels = image_input.get_file(tes_dir,False,0)
n_test = len(tes_labels)
tes_batch, label_batch = image_input.get_batch(tes_images, tes_labels,
BATCH_SIZE, False)
pre_logits = list_model.inference(tes_batch,
N_CLASSES,
is_train=is_training,
keep_prob=keep_prob)
logits = tool.FC_Layer('fc7', pre_logits, out_nodes=7)
y_pred = tf.argmax(logits, 1)
top_k_op = tf.nn.in_top_k(logits, label_batch, 1)
# class_pred = np.zeros(shape=n_test, dtype=np.int)
saver = tf.train.Saver(tf.global_variables())
with tf.Session() as sess:
print("Reading checkpoints...")
ckpt = tf.train.get_checkpoint_state(log_dir)
if ckpt and ckpt.model_checkpoint_path:
ckpt.model_checkpoint_path = 'D:\Leung\lenet5\log\log1017_cnn_c0.3\model.ckpt-6000'
global_step = ckpt.model_checkpoint_path.split('\\')[-1].split(
'-')[-1]
saver.restore(sess, ckpt.model_checkpoint_path)
print('Loading success, global_step is %s' % global_step)
else:
print('No checkpoint file found')
return
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
try:
num_iter = int(math.ceil(n_test / BATCH_SIZE))
true_count = 0
total_sample_count = num_iter * BATCH_SIZE
step = 0
# while step<1:
while step < num_iter and not coord.should_stop():
# class_test=np.append(class_test,tes_label)
# sess.run([tes_batch, label_batch])
t_label, logits_label, class_pred, predictions = sess.run(
[label_batch, logits, y_pred, top_k_op],
feed_dict={
is_training: False,
keep_prob: 1.0
})
# logits_label,class_pred = sess.run([logits, y_pred] )
pred_label = np.append(pred_label, class_pred)
true_label = np.append(true_label, t_label)
true_count += np.sum(predictions)
step += 1
precision = float(true_count) / total_sample_count
# print class_pred
print('precision = %.5f' % precision)
print(true_count, total_sample_count)
# print(pred_label)
for i in range(n_test):
if pred_label[i] == true_label[i]:
count += 1
print(count)
list_model.plot_confusion_matrix(pred_label, true_label)
except Exception as e:
coord.request_stop(e)
finally:
coord.request_stop()
coord.join(threads)
IMG_W = 64 # resize图像，太大的话训练时间久 IMG_H = 64 BATCH_SIZE = 20 CAPACITY = 200 MAX_STEP = 200 # 一般大于10K learning_rate = 0.0001 # 一般小于0.0001 # 获取批次batch train_dir = '/Users/gore/Desktop/object-detection/output' # 训练样本的读入路径 logs_train_dir = '/Users/gore/Desktop/object-detection/output' # logs存储路径 # logs_test_dir = 'E:/Re_train/image_data/test' #logs存储路径 # train, train_label = input_data.get_files(train_dir) train, train_label, val, val_label = image_input.get_files(train_dir, 0.3) # 训练数据及标签 train_batch, train_label_batch = image_input.get_batch(train, train_label, IMG_W, IMG_H, BATCH_SIZE, CAPACITY) # 测试数据及标签 val_batch, val_label_batch = image_input.get_batch(val, val_label, IMG_W, IMG_H, BATCH_SIZE, CAPACITY) # 训练操作定义 train_logits = cnn_model.inference(train_batch, BATCH_SIZE, N_CLASSES) train_loss = cnn_model.losses(train_logits, train_label_batch) train_op = cnn_model.trainning(train_loss, learning_rate) train_acc = cnn_model.evaluation(train_logits, train_label_batch) # 测试操作定义 test_logits = cnn_model.inference(val_batch, BATCH_SIZE, N_CLASSES) test_loss = cnn_model.losses(test_logits, val_label_batch) test_acc = cnn_model.evaluation(test_logits, val_label_batch) # 这个是log汇总记录