tensorflow框架----ckpt转pb模型

1、ckpt转pb

import tensorflow as tf

def freeze_graph(input_checkpoint, output_graph):
	# 指定输出的节点名称,该节点名称必须是原模型中存在的节点(重要！！！！！！！！！，如何找到该节点名称，见下文)
    output_node_names = ['attn_cell_1/transpose_1']  
    saver = tf.train.import_meta_graph(input_checkpoint + '.meta', clear_devices=True)
    graph = tf.get_default_graph()  # 获得默认的图
    input_graph_def = graph.as_graph_def()  # 返回一个序列化的图代表当前的图

    with tf.Session() as sess:
        saver.restore(sess, input_checkpoint)  # 恢复图并得到数据
        output_graph_def = tf.graph_util.convert_variables_to_constants(  # 模型持久化，将变量值固定
            sess=sess,
            input_graph_def=input_graph_def,  # 等于:sess.graph_def
            output_node_names=output_node_names)  # 如果有多个输出节点，以逗号隔开
        tensor_name_list = [tensor.name for tensor in input_graph_def.node]
        # print(tensor_name_list[:10])
        # sys.exit()

        with tf.gfile.FastGFile(output_graph, "wb") as f:  # 保存模型
            f.write(output_graph_def.SerializeToString())  # 序列化输出
        print("%d ops in the final graph." % len(output_graph_def.node))  # 得到当前图有几个操作节点
        for node in output_graph_def.node:
            print(node.name)

if __name__ == '__main__':
    # 输入ckpt模型路径
    input_checkpoint='results/full/model_final.weights/a'
    # 输出pb模型的路径
    out_pb_path="frozen_model.pb"
    # 调用freeze_graph将ckpt转为pb
    freeze_graph(input_checkpoint,out_pb_path)

2、pb模型预测

def freeze_graph_test(pb_path, image_list):
    '''
    :param pb_path:pb文件的路径
    :param image_list:测试图片的路径
    '''
    with tf.Graph().as_default():
        output_graph_def = tf.GraphDef()
        with open(pb_path, "rb") as f:
            output_graph_def.ParseFromString(f.read())
            tf.import_graph_def(output_graph_def, name="")
        with tf.Session() as sess:
            sess.run(tf.global_variables_initializer())

            # tensor_name_list = [tensor.name for tensor in output_graph_def.node]
            # print(tensor_name_list[:10])
 
            # 定义输入的张量名称,对应网络结构的输入张量
            # input:0作为输入图像,keep_prob:0作为dropout的参数,测试时值为1,is_training:0训练参数
            input_image_tensor = sess.graph.get_tensor_by_name("img:0")
            droupout_tensor = sess.graph.get_tensor_by_name("dropout:0")
        
            # 定义输出的张量名称
            output_tensor_name = sess.graph.get_tensor_by_name("attn_cell_1/transpose_1:0")

            tensor_list = [output_tensor_name,input_image_tensor,droupout_tensor]

            hypo_final = formulaRec(image_list, sess, tensor_list)

            print(hypo_final)

            str_out = convert_str(hypo_final[0])
            print(str_out)