程式碼：

import tensorflow as tf
 
sess = tf.Session()
check_point_path = 'variables' 
saver = tf.train.import_meta_graph('variables/save_variables.ckpt.meta')
 
saver.restore(sess,tf.train.latest_checkpoint(check_point_path))
 
graph = tf.get_default_graph()
 
#print(graph.get_operations())
 
#with open('op.txt','a') as f:
# f.write(str(graph.get_operations()))
op1 = graph.get_tensor_by_name('fully_connected/biases:0')
print(op1)

使用函式graph.get_operations()獲取ckpt.meta中儲存的graph中的所有operation，而tensor_name為'op_name:0'。

然後使用graph.get_tensor_by_name('op_name:0') 獲取tensor資訊。

程式碼從ckpt檔案中獲取儲存的variable的資料(tensor的name和value)：

import os
import tensorflow as tf
from tensorflow.python import pywrap_tensorflow
check_point_path = 'variables'
#checkpoint_path = os.path.join(logs_train_dir,'model.ckpt')
ckpt = tf.train.get_checkpoint_state(checkpoint_dir=check_point_path)
checkpoint_path = os.path.join('.',ckpt.model_checkpoint_path)
#print(ckpt.model_checkpoint_path)
reader = pywrap_tensorflow.NewCheckpointReader(checkpoint_path)
var_to_shape_map = reader.get_variable_to_shape_map()
for key in var_to_shape_map:
 print("tensor_name: ",key)
 #print(reader.get_tensor(key))

法二：

from tensorflow.python.tools.inspect_checkpoint import print_tensors_in_checkpoint_file
 
print_tensors_in_checkpoint_file("variables/save_variables.ckpt",tensor_name='',all_tensors=False,all_tensor_names=False)

注意：tf.train.latest_checkpoint(check_point_path) 方法用來獲取最後一次ckeckpoint的路徑，等價於

ckpt = tf.train.get_checkpoint_state(check_point_path)
ckpt.model_checkpoint_path

不能將tf.train.latest_checkpoint與tf.train.get_checkpoint_state 搞混了

以上這篇tensorflow 實現從checkpoint中獲取graph資訊就是小編分享給大家的全部內容了，希望能給大家一個參考，也希望大家多多支援我們。