有人说要 更换Keras版本降级，感觉或许有用，但是为了统一版本又找到了一个解决方案，姑且不只还有什么坑
import keras.backend as K
k.set_learning_phase(0)
解决方案是将学习阶段设置为测试模式。
调用方式
如何使用
Keras模型可以使用该功能保存为单个[ .hdf5或h5]文件，该文件同时存储体系结构和权重model.save()。然后可以通过如下调用此工具将该模型转换为TensorFlow模型：
python keras_to_tensorflow.py
--input_model="path/to/keras/model.h5"
--output_model="path/to/save/model.pb"
Keras模型也可以保存在两个单独的文件中，其中[ .hdf5或h5]文件使用该model.save_weights()功能存储权重，另一个.json文件使用该model.to_json()功能存储网络架构。在这种情况下，可以按以下方式转换模型：
python keras_to_tensorflow.py
--input_model="path/to/keras/model.h5"
--input_model_json="path/to/keras/model.json"
--output_model="path/to/save/model.pb"
尝试
python keras_to_tensorflow.py --help
了解其他受支持的标志（量化，output_nodes_prefix，save_graph_def）。
这是修改完的最终代码（从某github上download）
1 # -*- encoding: utf-8 -*- 2 """ 3 @File : ker_to_pb.py 4 @Time : 2020/5/12 20:23 5 @Author : xuluda 6 @Email : xuluda@163.com 7 @Software: PyCharm 8 9 """ 10 11 import tensorflow as tf 12 from tensorflow.python.framework import graph_util 13 from tensorflow.python.framework import graph_io 14 from pathlib import Path 15 from absl import app 16 from absl import flags 17 from absl import logging 18 import keras 19 from keras import backend as K 20 from keras.models import model_from_json, model_from_yaml 21 22 K.set_learning_phase(0) 23 FLAGS = flags.FLAGS 24 25 flags.DEFINE_string('input_model', None, 'Path to the input model.') 26 flags.DEFINE_string('input_model_json', None, 'Path to the input model ' 27 'architecture in json format.') 28 flags.DEFINE_string('input_model_yaml', None, 'Path to the input model ' 29 'architecture in yaml format.') 30 flags.DEFINE_string('output_model', None, 'Path where the converted model will ' 31 'be stored.') 32 flags.DEFINE_boolean('save_graph_def', False, 33 'Whether to save the graphdef.pbtxt file which contains ' 34 'the graph definition in ASCII format.') 35 flags.DEFINE_string('output_nodes_prefix', None, 36 'If set, the output nodes will be renamed to ' 37 '`output_nodes_prefix`+i, where `i` will numerate the ' 38 'number of of output nodes of the network.') 39 flags.DEFINE_boolean('quantize', False, 40 'If set, the resultant TensorFlow graph weights will be ' 41 'converted from float into eight-bit equivalents. See ' 42 'documentation here: ' 43 'https://github.com/tensorflow/tensorflow/tree/master/tensorflow/tools/graph_transforms') 44 flags.DEFINE_boolean('channels_first', False, 45 'Whether channels are the first dimension of a tensor. ' 46 'The default is TensorFlow behaviour where channels are ' 47 'the last dimension.') 48 flags.DEFINE_boolean('output_meta_ckpt', False, 49 'If set to True, exports the model as .meta, .index, and ' 50 '.data files, with a checkpoint file. These can be later ' 51 'loaded in TensorFlow to continue training.') 52 53 flags.mark_flag_as_required('input_model') 54 flags.mark_flag_as_required('output_model') 55 56 57 def load_model(input_model_path, input_json_path=None, input_yaml_path=None): 58 if not Path(input_model_path).exists(): 59 raise FileNotFoundError( 60 'Model file `{}` does not exist.'.format(input_model_path)) 61 try: 62 model = keras.models.load_model(input_model_path) 63 return model 64 except FileNotFoundError as err: 65 logging.error('Input mode file (%s) does not exist.', FLAGS.input_model) 66 raise err 67 except ValueError as wrong_file_err: 68 if input_json_path: 69 if not Path(input_json_path).exists(): 70 raise FileNotFoundError( 71 'Model description json file `{}` does not exist.'.format( 72 input_json_path)) 73 try: 74 model = model_from_json(open(str(input_json_path)).read()) 75 model.load_weights(input_model_path) 76 return model 77 except Exception as err: 78 logging.error("Couldn't load model from json.") 79 raise err 80 elif input_yaml_path: 81 if not Path(input_yaml_path).exists(): 82 raise FileNotFoundError( 83 'Model description yaml file `{}` does not exist.'.format( 84 input_yaml_path)) 85 try: 86 model = model_from_yaml(open(str(input_yaml_path)).read()) 87 model.load_weights(input_model_path) 88 return model 89 except Exception as err: 90 logging.error("Couldn't load model from yaml.") 91 raise err 92 else: 93 logging.error( 94 'Input file specified only holds the weights, and not ' 95 'the model definition. Save the model using ' 96 'model.save(filename.h5) which will contain the network ' 97 'architecture as well as its weights. ' 98 'If the model is saved using the ' 99 'model.save_weights(filename) function, either ' 100 'input_model_json or input_model_yaml flags should be set to ' 101 'to import the network architecture prior to loading the ' 102 'weights. \n' 103 'Check the keras documentation for more details ' 104 '(https://keras.io/getting-started/faq/)') 105 raise wrong_file_err 106 107 108 def main(args): 109 # If output_model path is relative and in cwd, make it absolute from root 110 output_model = FLAGS.output_model 111 if str(Path(output_model).parent) == '.': 112 output_model = str((Path.cwd() / output_model)) 113 114 output_fld = Path(output_model).parent 115 output_model_name = Path(output_model).name 116 output_model_stem = Path(output_model).stem 117 output_model_pbtxt_name = output_model_stem + '.pbtxt' 118 119 # Create output directory if it does not exist 120 Path(output_model).parent.mkdir(parents=True, exist_ok=True) 121 122 if FLAGS.channels_first: 123 K.set_image_data_format('channels_first') 124 else: 125 K.set_image_data_format('channels_last') 126 127 model = load_model(FLAGS.input_model, FLAGS.input_model_json, FLAGS.input_model_yaml) 128 129 # TODO(amirabdi): Support networks with multiple inputs 130 orig_output_node_names = [node.op.name for node in model.outputs] 131 if FLAGS.output_nodes_prefix: 132 num_output = len(orig_output_node_names) 133 pred = [None] * num_output 134 converted_output_node_names = [None] * num_output 135 136 # Create dummy tf nodes to rename output 137 for i in range(num_output): 138 converted_output_node_names[i] = '{}{}'.format( 139 FLAGS.output_nodes_prefix, i) 140 pred[i] = tf.identity(model.outputs[i], 141 name=converted_output_node_names[i]) 142 else: 143 converted_output_node_names = orig_output_node_names 144 logging.info('Converted output node names are: %s', 145 str(converted_output_node_names)) 146 147 sess = K.get_session() 148 if FLAGS.output_meta_ckpt: 149 saver = tf.train.Saver() 150 saver.save(sess, str(output_fld / output_model_stem)) 151 152 if FLAGS.save_graph_def: 153 tf.train.write_graph(sess.graph.as_graph_def(), str(output_fld), 154 output_model_pbtxt_name, as_text=True) 155 logging.info('Saved the graph definition in ascii format at %s', 156 str(Path(output_fld) / output_model_pbtxt_name)) 157 158 if FLAGS.quantize: 159 from tensorflow.tools.graph_transforms import TransformGraph 160 transforms = ["quantize_weights", "quantize_nodes"] 161 transformed_graph_def = TransformGraph(sess.graph.as_graph_def(), [], 162 converted_output_node_names, 163 transforms) 164 constant_graph = graph_util.convert_variables_to_constants( 165 sess, 166 transformed_graph_def, 167 converted_output_node_names) 168 else: 169 constant_graph = graph_util.convert_variables_to_constants( 170 sess, 171 sess.graph.as_graph_def(), 172 converted_output_node_names) 173 174 graph_io.write_graph(constant_graph, str(output_fld), output_model_name, 175 as_text=False) 176 logging.info('Saved the freezed graph at %s', 177 str(Path(output_fld) / output_model_name)) 178 179 180 if __name__ == "__main__": 181 app.run(main)
最终代码放一下（以下是第一次搞得代码 虽然能用 但是最后在openvino 上转换的时候 说我的pb 冻结的不完整）
# -*- encoding: utf-8 -*-
"""
@File : h5_to_pb.py
@Time : 2020/5/11 17:54
@Author : xuluda
@Email : xuluda@163.com
@Software: PyCharm
将keras的.h5的模型文件，转换成TensorFlow的pb文件
"""
# ==========================================================
from keras.models import load_model
import tensorflow as tf
import os.path as osp
import os
from keras import backend
backend.set_learning_phase(0)
# from keras.models import Sequential
def h5_to_pb(h5_model, output_dir, model_name, out_prefix="output_", log_tensorboard=True):
""".h5模型文件转换成pb模型文件
Argument:
h5_model: str
.h5模型文件
output_dir: str
pb模型文件保存路径
model_name: str
pb模型文件名称
out_prefix: str
根据训练，需要修改
log_tensorboard: bool
是否生成日志文件
Return:
pb模型文件
"""
if os.path.exists(output_dir) == False:
os.mkdir(output_dir)
out_nodes = []
for i in range(len(h5_model.outputs)):
out_nodes.append(out_prefix + str(i + 1))
tf.identity(h5_model.output[i], out_prefix + str(i + 1))
sess = backend.get_session()
from tensorflow.python.framework import graph_util, graph_io
# 写入pb模型文件
init_graph = sess.graph.as_graph_def()
main_graph = graph_util.convert_variables_to_constants(sess, init_graph, out_nodes)
graph_io.write_graph(main_graph, output_dir, name=model_name, as_text=False)
# 输出日志文件
if log_tensorboard:
from tensorflow.python.tools import import_pb_to_tensorboard
import_pb_to_tensorboard.import_to_tensorboard(os.path.join(output_dir, model_name), output_dir)
if __name__ == '__main__':
# .h模型文件路径参数
input_path = '/Users/jack/Documents/DaiCode/T4/'
weight_file = 'false_positive3.h5'
weight_file_path = os.path.join(input_path, weight_file)
output_graph_name = weight_file[:-3] + '.pb'
# pb模型文件输出输出路径
output_dir = osp.join(os.getcwd(), "trans_model")
# model.save(xingren.h5)
# 加载模型
# h5_model = Sequential()
h5_model = load_model(weight_file_path)
# h5_model.save(weight_file_path)
# h5_model.save('xingren.h5')
h5_to_pb(h5_model, output_dir=output_dir, model_name=output_graph_name)
print('Finished')