通过检查点的动量训练的模型的Tensorflow负载权重

Question

我正在尝试将使用Tensorflow训练的模型的权重加载到新的Keras模型中。

当我尝试评估变量以获取权重时，我得到以下信息：

with sess.as_default():
    # import graph
    saver = tf.train.import_meta_graph(meta_path)

    # load weights for graph
    saver.restore(sess, meta_path[:-5])

    # get all global variables (including model variables)
    vars_global = tf.global_variables()

    # get their name and value and put them into dictionary
    model_vars = {}
    for var in vars_global:
        try:
            model_vars[var.name] = var.eval()
        except Exception as e:
            print("For var={}, an exception occurred".format(var.name))
            print(e)

for key in sorted(model_vars.keys()):
    print(key, model_vars[key].shape)

global_step:0 ()
model/conv_1/biases:0 (128,)
model/conv_1/weights:0 (8, 500, 1, 128)
model/conv_2/biases:0 (128,)
model/conv_2/weights:0 (8, 500, 1, 128)
model/embed/embedding_vars:0 (257, 8)
model/fc_1/biases:0 (128,)
model/fc_1/weights:0 (128, 128)
model/fc_2/biases:0 (1,)
model/fc_2/weights:0 (128, 1)
model/model/conv_1/biases/Momentum:0 (128,)
model/model/conv_1/weights/Momentum:0 (8, 500, 1, 128)
model/model/conv_2/biases/Momentum:0 (128,)
model/model/conv_2/weights/Momentum:0 (8, 500, 1, 128)
model/model/embed/embedding_vars/Momentum:0 (257, 8)
model/model/fc_1/biases/Momentum:0 (128,)
model/model/fc_1/weights/Momentum:0 (128, 128)
model/model/fc_2/biases/Momentum:0 (1,)
model/model/fc_2/weights/Momentum:0 (128, 1)

我应该在新模型的各层中使用哪两组不同的权重？还是应该以某种方式将它们结合起来？

我尝试过分别加载两者，但没有得到任何有意义的结果。