tensorflow保存的预训练的模型包含3个文件:
- model_name.data-00000-of-00001 (保存网络中每个变量值,包括所有的weights、biases、gradients等变量, 类似python字典结果,对应每个变量的数值(key,value)。)
- model_name.index (是对应模型的索引文件)
- model_name.meta (文件保存的是图结构,meta文件是pb(protocol buffer)格式文件,包含变量、op、集合等。)
打印保存在model里的变量名和变量值:
from tensorflow.python import pywrap_tensorflow
checkpoint_path = "./*/model_name"
reader = pywrap_tensorflow.NewCheckpointReader(checkpoint_path)
vars = reader.get_variable_to_shape_map()
for v in vars:
v_name = v
var_tensor = reader.get_tensor(v)
print("tensor name = ",v)
print("tensor shape = ",var_tensor.shape)
print("tensor value = ", var_tensor)