Tensorflow 1.0之后模型文件、权重数值的读取方式

保存的文件有4个：

checkpoint

model-parameters.bin-46000.data-00000-of-00001

model-parameters.bin-46000.index

model-parameters.bin-46000.meta

读取代码：

from tensorflow.python import pywrap_tensorflow

with tf.Graph().as_default():
with tf.Session() as sess:
ckpt = tf.train.get_checkpoint_state('./checkpoint')
if ckpt and ckpt.model_checkpoint_path:
 reader = pywrap_tensorflow.NewCheckpointReader('./checkpoint/model-parameters.bin-46000')
 all_variables = reader.get_variable_to_shape_map()
 w1 = reader.get_tensor("conv1/weight")
 print(w1.shape)
 print(w1)
 print(all_variables)
else:
 print('No checkpoint file found')

来源：https://blog.csdn.net/liujianlin01/article/details/80426348