pt模型转rknn后识别率变差

转换模型的代码很多，列举一二如下：
一：直接修改就完事儿，只要你能找对输出节点

"""
    -通过传入 CKPT 模型的路径得到模型的图和变量数据
    -通过 import_meta_graph 导入模型中的图
    -通过 saver.restore 从模型中恢复图中各个变量的数据
    -通过 graph_util.convert_variables_to_constants 将模型持久化
"""
 
import tensorflow as tf
from create_tf_record import *
from tensorflow.python.framework import graph_util
 
resize_height = 299  # 指定图片高度
resize_width = 299  # 指定图片宽度
depths = 3
 
def freeze_graph_test(pb_path, image_path):
    '''
    :param pb_path:pb文件的路径
    :param image_path:测试图片的路径
    :return:
    '''
    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())
 
            # 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("input:0")
            input_keep_prob_tensor = sess.graph.get_tensor_by_name("keep_prob:0")
            input_is_training_tensor = sess.graph.get_tensor_by_name("is_training:0")
 
            # 定义输出的张量名称
            output_tensor_name = sess.graph.get_tensor_by_name("InceptionV3/Logits/SpatialSqueeze:0")
 
            # 读取测试图片
            im=read_image(image_path,resize_height,resize_width,normalization=True)
            im=im[np.newaxis,:]
            # 测试读出来的模型是否正确，注意这里传入的是输出和输入节点的tensor的名字，不是操作节点的名字
            # out=sess.run("InceptionV3/Logits/SpatialSqueeze:0", feed_dict={'input:0': im,'keep_prob:0':1.0,'is_training:0':False})
            out=sess.run(output_tensor_name, feed_dict={input_image_tensor: im,
                                                        input_keep_prob_tensor:1.0,
                                                        input_is_training_tensor:False})
            print("out:{}".format(out))
            score = tf.nn.softmax(out, name='pre')
            class_id = tf.argmax(score, 1)
            print "pre class_id:{}".format(sess.run(class_id))
 
 
def freeze_graph(input_checkpoint,output_graph):
    '''
    :param input_checkpoint:
    :param output_graph: PB模型保存路径
    :return:
    '''
    # checkpoint = tf.train.get_checkpoint_state(model_folder) #检查目录下ckpt文件状态是否可用
    # input_checkpoint = checkpoint.model_checkpoint_path #得ckpt文件路径
 
    # 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
    output_node_names = "InceptionV3/Logits/SpatialSqueeze"
    saver = tf.train.import_meta_graph(input_checkpoint + '.meta', clear_devices=True)
 
    with tf.Session() as sess:
        saver.restore(sess, input_checkpoint) #恢复图并得到数据
        output_graph_def = graph_util.convert_variables_to_constants(  # 模型持久化，将变量值固定
            sess=sess,
            input_graph_def=sess.graph_def,# 等于:sess.graph_def
            output_node_names=output_node_names.split(","))# 如果有多个输出节点，以逗号隔开
 
        with tf.gfile.GFile(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 op in sess.graph.get_operations():
        #     print(op.name, op.values())
 
def freeze_graph2(input_checkpoint,output_graph):
    '''
    :param input_checkpoint:
    :param output_graph: PB模型保存路径
    :return:
    '''
    # checkpoint = tf.train.get_checkpoint_state(model_folder) #检查目录下ckpt文件状态是否可用
    # input_checkpoint = checkpoint.model_checkpoint_path #得ckpt文件路径
 
    # 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
    output_node_names = "InceptionV3/Logits/SpatialSqueeze"
    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 = 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.split(","))# 如果有多个输出节点，以逗号隔开
 
        with tf.gfile.GFile(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 op in graph.get_operations():
        #     print(op.name, op.values())
 
 
if __name__ == '__main__':
    # 输入ckpt模型路径
    input_checkpoint='models/model.ckpt-10000'
    # 输出pb模型的路径
    out_pb_path="models/pb/frozen_model.pb"
    # 调用freeze_graph将ckpt转为pb
    freeze_graph(input_checkpoint,out_pb_path)
 
    # 测试pb模型
    image_path = 'test_image/animal.jpg'
    freeze_graph_test(pb_path=out_pb_path, image_path=image_path)

二，通过demo测试加入节点转换成pb模型1.0

graph = tf.get_default_graph()
     input_graph_def = graph.as_graph_def()
     output_graph = "vgg16_frozen_model.pb"
     output_node_names = "vgg16/cls_pred,vgg_16/rpn_bbox_pred/biases,vgg_16_1/rois/Reshape,vgg_16/rpn_cls_score/biases"
     output_graph_def = graph_util.convert_variables_to_constants(sess, input_graph_def, output_node_names.split(","))
     with tf.gfile.GFile(output_graph, "wb") as f:
         f.write(output_graph_def.SerializeToString())

三，通过demo测试加入节点转换成pb模型2.0

# 保存图
     tf.train.write_graph(sess.graph_def, 'pb/pb_model', 'model.pb')
     # 把图和参数结构一起
     freeze_graph.freeze_graph('pb/pb_model/model.pb',
                               '',
                               False,
                               tfmodel,
                               'vgg_16/rpn_cls_score/biases,vgg_16/cls_pred,vgg_16/rpn_bbox_pred/biases,vgg_16/rois/PyFunc',
                               'save/restore_all',
                               'save/Const:0',
                               'pb/pb_model/vgg16_frozen_model.pb',
                              False,
                               "")