tensorflow 获取checkpoint中的变量列表

方式1:静态获取,通过直接解析checkpoint文件获取变量名及变量值

通过

reader = tf.train.NewCheckpointReader(model_path)

或者通过:

from tensorflow.python import pywrap_tensorflow
reader = pywrap_tensorflow.NewCheckpointReader(model_path)

代码:

model_path = "./checkpoints/model.ckpt-75000"
## 下面两个reader作用等价
#reader = pywrap_tensorflow.NewCheckpointReader(model_path)
reader = tf.train.NewCheckpointReader(model_path)

## 用reader获取变量字典,key是变量名,value是变量的shape
var_to_shape_map = reader.get_variable_to_shape_map()
for var_name in var_to_shape_map.keys():
    #用reader获取变量值
    var_value = reader.get_tensor(var_name)

    print("var_name",var_name)
    print("var_value",var_value)

方式2:动态获取,先加载checkpoint模型,然后用graph.get_tensor_by_name()获取变量值

代码(注意:要先在脚本中构建model中对应的变量及scope):

    model_path = "./checkpoints/model.ckpt-75000"
    config = tf.ConfigProto()
    config.gpu_options.allow_growth = True
    with tf.Session(config=config) as sess:
        ## 获取待加载的变量列表
        trainable_vars = tf.trainable_variables()
        g_vars = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES,scope="generator")
        d_vars = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES,scope='discriminator')
        flow_vars = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES,scope='flow_net')
        var_restore = g_vars + d_vars

        ## 仅加载目标变量
        loader = tf.train.Saver(var_restore)
        loader.restore(sess,model_path)

        ## 显示加载的变量值
        graph = tf.get_default_graph()
        for var in var_restore:
            tensor = graph.get_tensor_by_name(var.name)
            print("=======变量名=======",tensor)
            print("-------变量值-------",sess.run(tensor))