TensorFlow实现模型断点训练,checkpoint模型载入方式

# 保存模型
saver = tf.train.Saver(max_to_keep=1) # 最多保留最新的模型
 
# 开启会话
with tf.Session() as sess:
 # saver.restore(sess, './log/' + "model_savemodel.cpkt-" + str(20000))
 sess.run(tf.global_variables_initializer())
 ckpt = tf.train.get_checkpoint_state('./log/') # 注意此处是checkpoint存在的目录，千万不要写成‘./log'
 if ckpt and ckpt.model_checkpoint_path:
 saver.restore(sess,ckpt.model_checkpoint_path) # 自动恢复model_checkpoint_path保存模型一般是最新
 print("Model restored...")
 else:
 print('No Model')

# 保存模型
saver = tf.train.Saver(max_to_keep=1)
# 开启会话
 
with tf.Session() as sess:
 saver.restore(sess, './log/' + "model_savemodel.cpkt-" + str(111000))
 sess.run(tf.global_variables_initializer())

Model restored...
Iter:0, D_loss:0.5139875411987305, G_loss:2.8023970127105713
Iter:1000, D_loss:0.4400891065597534, G_loss:2.781547784805298
Iter:2000, D_loss:0.5169454216957092, G_loss:2.58009934425354
Iter:3000, D_loss:0.4507023096084595, G_loss:2.584151268005371
Iter:4000, D_loss:0.5746167898178101, G_loss:2.5365757942199707
Iter:5000, D_loss:0.5288565158843994, G_loss:2.426676034927368
Iter:6000, D_loss:0.549595057964325, G_loss:2.820535659790039
Iter:7000, D_loss:0.32620012760162354, G_loss:2.540236473083496
Iter:8000, D_loss:0.4363398551940918, G_loss:2.5880446434020996
Iter:9000, D_loss:0.569464921951294, G_loss:2.5133447647094727
done！

import tensorflow as tf

graph=tf.get_default_graph()  # 获取当前图
sess=tf.Session()
sess.run(tf.global_variables_initializer())

checkpoint_file=tf.train.latest_checkpoint(model_path)
saver = tf.train.import_meta_graph("{}.meta".format(checkpoint_file))
saver.restore(sess,checkpoint_file)

import tensorflow as tf

graph=tf.get_default_graph()  # 获取当前图
sess=tf.Session()
sess.run(tf.global_variables_initializer())

checkpoint_file=os.path.join(model_path,'model-9200')
saver = tf.train.import_meta_graph("{}.meta".format(checkpoint_file))
saver.restore(sess,checkpoint_file)

## 得到当前图中所有变量的名称
tensor_name_list=[tensor.name for tensor in graph.as_graph_def().node] 
# 查看所有变量
print(tensor_name_list) 

# 获取input_x和input_y的变量值
input_x = graph.get_operation_by_name("input_x").outputs[0]
input_y = graph.get_operation_by_name("input_y").outputs[0]