if __name__ == '__main__':
images, labels = create_inputs_seg_hand(is_train=is_train)
session_config = tf.ConfigProto(
device_count={'GPU': 0},
gpu_options={
'allow_growth': 1,
# 'per_process_gpu_memory_fraction': 0.1,
'visible_device_list': '0'
},
allow_soft_placement=True) ##这个设置必须有，否则无论如何都会报cudnn不匹配的错误,BUG十分隐蔽，真是智障
with tf.Session(config=session_config) as sess:
# 1、先定义model才能执行第二步的初始化
model = SegCaps(sess, cfg, is_train=is_train)
# 2、初始化和启动线程
tf.global_variables_initializer().run()
tf.local_variables_initializer().run()
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
model.restore(model_restore_name)
#列表初始化
AUC_ROC_list,AUC_prec_rec_list,accuracy_list,specificity_list,sensitivity_list,\
precision_list,jaccard_index_list,F1_score_list,all_list,mean_list\
=[],[],[],[],[],[],[],[],[],[]
#3、测试图片
index = 0
for i in range(test_data_number // batch_size):
from v1.use_seg_tfrecord import create_inputs_seg_hand
if __name__ == '__main__':
images, labels = create_inputs_seg_hand(is_train=True)
session_config = tf.ConfigProto(
device_count={'GPU': 0},
gpu_options={
'allow_growth': 1,
# 'per_process_gpu_memory_fraction': 0.1,
'visible_device_list': '0'
},
allow_soft_placement=True) ##这个设置必须有，否则无论如何都会报cudnn不匹配的错误,BUG十分隐蔽，真是智障
with tf.Session(config=session_config) as sess:
# 1、先定义model才能执行第二步的初始化
model = SegCaps(sess, cfg, is_train=True)
# 2、初始化和启动线程
tf.global_variables_initializer().run()
tf.local_variables_initializer().run()
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
# 3、训练模型
for i in range(num_epochs):
pics, pics_masks = sess.run([images, labels]) # 取出一个batchsize的图片
loss_value = model.fit(pics, pics_masks, summary_step=i)
print('loss_value: ', loss_value)
model.save()
coord.request_stop()
coord.join(threads)
if __name__ == '__main__':
images, labels = create_inputs_seg_hand(is_train=True)
session_config = tf.ConfigProto(
device_count={'GPU': 0},
gpu_options={
'allow_growth': 1,
# 'per_process_gpu_memory_fraction': 0.1,
'visible_device_list': '0'
},
allow_soft_placement=True) ##这个设置必须有，否则无论如何都会报cudnn不匹配的错误,BUG十分隐蔽，真是智障
with tf.Session(config=session_config) as sess:
# 1、先定义model才能执行第二步的初始化
model = SegCaps(sess, cfg, is_train=True)
# 2、初始化和启动线程
tf.global_variables_initializer().run()
tf.local_variables_initializer().run()
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
if model_restore_name:
model.restore(model_restore_name)
# 3、训练模型
# num_epochs=10000
for i in range(start_epoch, num_epochs):
since = time.time()
#1、读图
pics, pics_masks = sess.run([images, labels]) # 取出一个batchsize的图片
def start_eval():
images, labels = create_inputs_seg_hand(is_train=is_train)
session_config = tf.ConfigProto(
device_count={'GPU': 0},
gpu_options={
'allow_growth': 1,
# 'per_process_gpu_memory_fraction': 0.1,
'visible_device_list': '0'
},
allow_soft_placement=True) ##这个设置必须有，否则无论如何都会报cudnn不匹配的错误,BUG十分隐蔽，真是智障
with tf.Session(config=session_config) as sess:
# 1、先定义model才能执行第二步的初始化
model = SegCaps(sess, cfg, is_train=is_train)
# 2、初始化和启动线程
tf.global_variables_initializer().run()
tf.local_variables_initializer().run()
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
model.restore()
#列表初始化
AUC_ROC_list,AUC_prec_rec_list,accuracy_list,specificity_list,sensitivity_list,\
precision_list,jaccard_index_list,F1_score_list,all_list,mean_list\
=[],[],[],[],[],[],[],[],[],[]
#3、测试图片
index = 0
for i in range(test_data_number // batch_size):
pics, pics_masks = sess.run([images, labels]) # 取出一个batchsize的图片
# 3、计算耗时
since = time.time()
pre = model.predict(pics)
seconds = time.time() - since
pre_list = np.split(pre, batch_size, axis=0)
pres = np.squeeze(pre_list, axis=0)
for label, pre in zip(pics_masks, pres):
y_scores = pre.reshape(-1, 1)
y_true = label.reshape(-1, 1)
# 1、画ROC曲线
AUC_ROC = plot_roc_curve(y_true, y_scores)
AUC_ROC_list.append(AUC_ROC)
#2、画P_R-curve曲线
AUC_prec_rec = plot_precision_recall_curve(y_true, y_scores)
AUC_prec_rec_list.append(AUC_prec_rec)
#3、Confusion matrix
y_pred_binary = convert_to_binary(shape=y_scores.shape[0],
y_scores=y_scores)
accuracy, specificity, sensitivity, precision \
= plot_confusion_matrix(y_true, y_pred_binary)
accuracy_list.append(accuracy)
specificity_list.append(specificity)
sensitivity_list.append(sensitivity)
precision_list.append(precision)
#4、Jaccard similarity index
jaccard_index = get_jaccard_index(y_true, y_pred_binary)
jaccard_index_list.append(jaccard_index)
#5、F1 score
F1_score = get_F1_score(y_true, y_pred_binary)
F1_score_list.append(F1_score)
print(
'######################### end ####################################'
)
#1、评估数据存进列表中
all_list = [AUC_ROC_list, AUC_prec_rec_list, accuracy_list, specificity_list \
, sensitivity_list, precision_list, jaccard_index_list, F1_score_list]
name_list = [
'AUC_ROC', 'AUC_prec_rec', 'accuracy', 'specificity',
'sensitivity', 'precision', 'jaccard_index', 'F1_score'
]
# 2、panda保存所有图片的评估值到CSV文件
save_all_pics_value(name_list, all_list)
#3、panda保存平均值到CSV文件
save_mean_value(name_list, all_list)
#4、结束
coord.request_stop()
coord.join(threads)
if __name__ == '__main__':
images, labels = create_inputs_seg_hand(is_train=is_train)
session_config = tf.ConfigProto(
device_count={'GPU': 0},
gpu_options={
'allow_growth': 1,
# 'per_process_gpu_memory_fraction': 0.1,
'visible_device_list': '0'
},
allow_soft_placement=True) ##这个设置必须有，否则无论如何都会报cudnn不匹配的错误,BUG十分隐蔽，真是智障
with tf.Session(config=session_config) as sess:
# 1、先定义model才能执行第二步的初始化
model = SegCaps(sess, cfg, is_train=is_train)
# 2、初始化和启动线程
tf.global_variables_initializer().run()
tf.local_variables_initializer().run()
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(sess=sess, coord=coord)
model.restore()
#列表初始化
AUC_ROC_list,AUC_prec_rec_list,accuracy_list,specificity_list,sensitivity_list,\
precision_list,jaccard_index_list,F1_score_list,all_list,mean_list\
=[],[],[],[],[],[],[],[],[],[]
#3、测试图片
index = 0
for i in range(test_data_number // batch_size):