我用keras（tensorflow后端）训练了一个网络，并将模型另存为json，权重另存为h5。我现在正在尝试将其转换为单个tensorflow pb文件，并且它抱怨输出节点的名称。
系统信息： Tensorflow 2.3.0 凯拉斯2.4.3 CUDA 10.1 卡登7
转换脚本非常简单：
import json
from tensorflow import keras
from keras import backend as K
import tensorflow as tf
json_file = "my-trained-model.json"
h5_file = "my-trained-model.h5"
Output_Path = "./trained_models/"
Frozen_pb_File = "my-trained-model.pb"
def freeze_session(session, keep_var_names=None, output_names=None, clear_devices=True):
from tensorflow.python.framework.graph_util import convert_variables_to_constants
graph = session.graph
with graph.as_default():
freeze_var_names = list(set(v.op.name for v in tf.compat.v1.global_variables()).difference(keep_var_names or []))
output_names = output_names or []
output_names += [v.op.name for v in tf.compat.v1.global_variables()]
# Graph -> GraphDef ProtoBuf
input_graph_def = graph.as_graph_def()
if clear_devices:
for node in input_graph_def.node:
node.device = ""
frozen_graph = convert_variables_to_constants(session, input_graph_def,
output_names, freeze_var_names)
return frozen_graph
with open(json_file, 'r') as json_file:
model = keras.models.model_from_json(json_file.read())
model.load_weights(h5_file)
model.summary()
# get output node names
OutputNames = [out.op.name for out in model.outputs]
print("\nOutput Names:\n", OutputNames) # this prints "concatenate/concat" as the only output node name
# freeze the model
frozen_graph = freeze_session(tf.compat.v1.keras.backend.get_session(), output_names=OutputNames)
# save the output files
# this is the .pb file (a binary file)
tf.io.write_graph(frozen_graph, Output_Path, Frozen_pb_File, as_text=False)
运行此命令时，
AssertionError: concatenate/concat is not in graph
因此由于某种原因，它正在读取“ concatenate / concat”的输出节点名称。下面给出了模型摘要，您可以看到输出节点是“ concatenate”的。但是，即使我将输出节点名称硬编码为“ concatenate”，我也会收到类似的断言错误：
AssertionError: concatenate is not in graph
这是keras模型的摘要：
Layer (type) Output Shape Param # Connected to
==================================================================================================
input (InputLayer) [(None, None, None, 0
__________________________________________________________________________________________________
conv2d (Conv2D) (None, None, None, 1 448 input[0][0]
__________________________________________________________________________________________________
batch_normalization (BatchNorma (None, None, None, 1 64 conv2d[0][0]
__________________________________________________________________________________________________
activation (Activation) (None, None, None, 1 0 batch_normalization[0][0]
__________________________________________________________________________________________________
conv2d_1 (Conv2D) (None, None, None, 1 2320 activation[0][0]
__________________________________________________________________________________________________
batch_normalization_1 (BatchNor (None, None, None, 1 64 conv2d_1[0][0]
__________________________________________________________________________________________________
activation_1 (Activation) (None, None, None, 1 0 batch_normalization_1[0][0]
__________________________________________________________________________________________________
max_pooling2d (MaxPooling2D) (None, None, None, 1 0 activation_1[0][0]
__________________________________________________________________________________________________
conv2d_2 (Conv2D) (None, None, None, 3 4640 max_pooling2d[0][0]
__________________________________________________________________________________________________
batch_normalization_2 (BatchNor (None, None, None, 3 128 conv2d_2[0][0]
__________________________________________________________________________________________________
activation_2 (Activation) (None, None, None, 3 0 batch_normalization_2[0][0]
__________________________________________________________________________________________________
conv2d_3 (Conv2D) (None, None, None, 3 9248 activation_2[0][0]
__________________________________________________________________________________________________
batch_normalization_3 (BatchNor (None, None, None, 3 128 conv2d_3[0][0]
__________________________________________________________________________________________________
activation_3 (Activation) (None, None, None, 3 0 batch_normalization_3[0][0]
__________________________________________________________________________________________________
conv2d_4 (Conv2D) (None, None, None, 3 9248 activation_3[0][0]
__________________________________________________________________________________________________
batch_normalization_4 (BatchNor (None, None, None, 3 128 conv2d_4[0][0]
__________________________________________________________________________________________________
add (Add) (None, None, None, 3 0 batch_normalization_4[0][0]
activation_2[0][0]
__________________________________________________________________________________________________
activation_4 (Activation) (None, None, None, 3 0 add[0][0]
__________________________________________________________________________________________________
max_pooling2d_1 (MaxPooling2D) (None, None, None, 3 0 activation_4[0][0]
__________________________________________________________________________________________________
conv2d_5 (Conv2D) (None, None, None, 6 18496 max_pooling2d_1[0][0]
__________________________________________________________________________________________________
batch_normalization_5 (BatchNor (None, None, None, 6 256 conv2d_5[0][0]
__________________________________________________________________________________________________
activation_5 (Activation) (None, None, None, 6 0 batch_normalization_5[0][0]
__________________________________________________________________________________________________
conv2d_6 (Conv2D) (None, None, None, 6 36928 activation_5[0][0]
__________________________________________________________________________________________________
batch_normalization_6 (BatchNor (None, None, None, 6 256 conv2d_6[0][0]
__________________________________________________________________________________________________
activation_6 (Activation) (None, None, None, 6 0 batch_normalization_6[0][0]
__________________________________________________________________________________________________
conv2d_7 (Conv2D) (None, None, None, 6 36928 activation_6[0][0]
__________________________________________________________________________________________________
batch_normalization_7 (BatchNor (None, None, None, 6 256 conv2d_7[0][0]
__________________________________________________________________________________________________
add_1 (Add) (None, None, None, 6 0 batch_normalization_7[0][0]
activation_5[0][0]
__________________________________________________________________________________________________
activation_7 (Activation) (None, None, None, 6 0 add_1[0][0]
__________________________________________________________________________________________________
conv2d_8 (Conv2D) (None, None, None, 6 36928 activation_7[0][0]
__________________________________________________________________________________________________
batch_normalization_8 (BatchNor (None, None, None, 6 256 conv2d_8[0][0]
__________________________________________________________________________________________________
activation_8 (Activation) (None, None, None, 6 0 batch_normalization_8[0][0]
__________________________________________________________________________________________________
conv2d_9 (Conv2D) (None, None, None, 6 36928 activation_8[0][0]
__________________________________________________________________________________________________
batch_normalization_9 (BatchNor (None, None, None, 6 256 conv2d_9[0][0]
__________________________________________________________________________________________________
add_2 (Add) (None, None, None, 6 0 batch_normalization_9[0][0]
activation_7[0][0]
__________________________________________________________________________________________________
activation_9 (Activation) (None, None, None, 6 0 add_2[0][0]
__________________________________________________________________________________________________
max_pooling2d_2 (MaxPooling2D) (None, None, None, 6 0 activation_9[0][0]
__________________________________________________________________________________________________
conv2d_10 (Conv2D) (None, None, None, 6 36928 max_pooling2d_2[0][0]
__________________________________________________________________________________________________
batch_normalization_10 (BatchNo (None, None, None, 6 256 conv2d_10[0][0]
__________________________________________________________________________________________________
activation_10 (Activation) (None, None, None, 6 0 batch_normalization_10[0][0]
__________________________________________________________________________________________________
conv2d_11 (Conv2D) (None, None, None, 6 36928 activation_10[0][0]
__________________________________________________________________________________________________
batch_normalization_11 (BatchNo (None, None, None, 6 256 conv2d_11[0][0]
__________________________________________________________________________________________________
activation_11 (Activation) (None, None, None, 6 0 batch_normalization_11[0][0]
__________________________________________________________________________________________________
conv2d_12 (Conv2D) (None, None, None, 6 36928 activation_11[0][0]
__________________________________________________________________________________________________
batch_normalization_12 (BatchNo (None, None, None, 6 256 conv2d_12[0][0]
__________________________________________________________________________________________________
add_3 (Add) (None, None, None, 6 0 batch_normalization_12[0][0]
activation_10[0][0]
__________________________________________________________________________________________________
activation_12 (Activation) (None, None, None, 6 0 add_3[0][0]
__________________________________________________________________________________________________
conv2d_13 (Conv2D) (None, None, None, 6 36928 activation_12[0][0]
__________________________________________________________________________________________________
batch_normalization_13 (BatchNo (None, None, None, 6 256 conv2d_13[0][0]
__________________________________________________________________________________________________
activation_13 (Activation) (None, None, None, 6 0 batch_normalization_13[0][0]
__________________________________________________________________________________________________
conv2d_14 (Conv2D) (None, None, None, 6 36928 activation_13[0][0]
__________________________________________________________________________________________________
batch_normalization_14 (BatchNo (None, None, None, 6 256 conv2d_14[0][0]
__________________________________________________________________________________________________
add_4 (Add) (None, None, None, 6 0 batch_normalization_14[0][0]
activation_12[0][0]
__________________________________________________________________________________________________
activation_14 (Activation) (None, None, None, 6 0 add_4[0][0]
__________________________________________________________________________________________________
max_pooling2d_3 (MaxPooling2D) (None, None, None, 6 0 activation_14[0][0]
__________________________________________________________________________________________________
conv2d_15 (Conv2D) (None, None, None, 1 73856 max_pooling2d_3[0][0]
__________________________________________________________________________________________________
batch_normalization_15 (BatchNo (None, None, None, 1 512 conv2d_15[0][0]
__________________________________________________________________________________________________
activation_15 (Activation) (None, None, None, 1 0 batch_normalization_15[0][0]
__________________________________________________________________________________________________
conv2d_16 (Conv2D) (None, None, None, 1 147584 activation_15[0][0]
__________________________________________________________________________________________________
batch_normalization_16 (BatchNo (None, None, None, 1 512 conv2d_16[0][0]
__________________________________________________________________________________________________
activation_16 (Activation) (None, None, None, 1 0 batch_normalization_16[0][0]
__________________________________________________________________________________________________
conv2d_17 (Conv2D) (None, None, None, 1 147584 activation_16[0][0]
__________________________________________________________________________________________________
batch_normalization_17 (BatchNo (None, None, None, 1 512 conv2d_17[0][0]
__________________________________________________________________________________________________
add_5 (Add) (None, None, None, 1 0 batch_normalization_17[0][0]
activation_15[0][0]
__________________________________________________________________________________________________
activation_17 (Activation) (None, None, None, 1 0 add_5[0][0]
__________________________________________________________________________________________________
conv2d_18 (Conv2D) (None, None, None, 1 147584 activation_17[0][0]
__________________________________________________________________________________________________
batch_normalization_18 (BatchNo (None, None, None, 1 512 conv2d_18[0][0]
__________________________________________________________________________________________________
activation_18 (Activation) (None, None, None, 1 0 batch_normalization_18[0][0]
__________________________________________________________________________________________________
conv2d_19 (Conv2D) (None, None, None, 1 147584 activation_18[0][0]
__________________________________________________________________________________________________
batch_normalization_19 (BatchNo (None, None, None, 1 512 conv2d_19[0][0]
__________________________________________________________________________________________________
add_6 (Add) (None, None, None, 1 0 batch_normalization_19[0][0]
activation_17[0][0]
__________________________________________________________________________________________________
activation_19 (Activation) (None, None, None, 1 0 add_6[0][0]
__________________________________________________________________________________________________
conv2d_20 (Conv2D) (None, None, None, 1 147584 activation_19[0][0]
__________________________________________________________________________________________________
batch_normalization_20 (BatchNo (None, None, None, 1 512 conv2d_20[0][0]
__________________________________________________________________________________________________
activation_20 (Activation) (None, None, None, 1 0 batch_normalization_20[0][0]
__________________________________________________________________________________________________
conv2d_21 (Conv2D) (None, None, None, 1 147584 activation_20[0][0]
__________________________________________________________________________________________________
batch_normalization_21 (BatchNo (None, None, None, 1 512 conv2d_21[0][0]
__________________________________________________________________________________________________
add_7 (Add) (None, None, None, 1 0 batch_normalization_21[0][0]
activation_19[0][0]
__________________________________________________________________________________________________
activation_21 (Activation) (None, None, None, 1 0 add_7[0][0]
__________________________________________________________________________________________________
conv2d_22 (Conv2D) (None, None, None, 1 147584 activation_21[0][0]
__________________________________________________________________________________________________
batch_normalization_22 (BatchNo (None, None, None, 1 512 conv2d_22[0][0]
__________________________________________________________________________________________________
activation_22 (Activation) (None, None, None, 1 0 batch_normalization_22[0][0]
__________________________________________________________________________________________________
conv2d_23 (Conv2D) (None, None, None, 1 147584 activation_22[0][0]
__________________________________________________________________________________________________
batch_normalization_23 (BatchNo (None, None, None, 1 512 conv2d_23[0][0]
__________________________________________________________________________________________________
add_8 (Add) (None, None, None, 1 0 batch_normalization_23[0][0]
activation_21[0][0]
__________________________________________________________________________________________________
activation_23 (Activation) (None, None, None, 1 0 add_8[0][0]
__________________________________________________________________________________________________
conv2d_24 (Conv2D) (None, None, None, 2 2306 activation_23[0][0]
__________________________________________________________________________________________________
conv2d_25 (Conv2D) (None, None, None, 6 6918 activation_23[0][0]
__________________________________________________________________________________________________
concatenate (Concatenate) (None, None, None, 8 0 conv2d_24[0][0]
conv2d_25[0][0]
==================================================================================================
Total params: 1,648,184
Trainable params: 1,644,344
Non-trainable params: 3,840
__________________________________________________________________________________________________
我必须俯瞰一些简单的事物，但是凝视了这么久，我再也看不到树木茂密的森林了。 ：-（
谢谢您的任何建议。我会很乐意与任何有兴趣的人分享json / h5。
答案 0 :(得分：0)
看起来这一切都是由于尝试冻结tensorflow 2.3模型引起的。显然，Tensorflow 2.0+已弃用“冻结”概念，并转移到“保存模型”概念。一旦发现这一点，我便能够立即将h5 / json保存到已保存的模型pb中。
我仍然不确定该格式是否针对推论进行了优化，因此我将对此进行一些跟进，但是由于我的问题是关于我所看到的错误的信息，我以为我会发布导致该错误的原因。问题。
作为参考，这是我的python脚本，用于将keras h5 / json文件转换为Tensorflow保存的模型格式。
import os
from keras.models import model_from_json
import tensorflow as tf
import genericpath
from genericpath import *
def splitext(p):
p = os.fspath(p)
if isinstance(p, bytes):
sep = b'/'
extsep = b'.'
else:
sep = '/'
extsep = '.'
return genericpath._splitext(p, sep, None, extsep)
def load_model(path,custom_objects={},verbose=0):
from keras.models import model_from_json
path = splitext(path)[0]
with open('%s.json' % path,'r') as json_file:
model_json = json_file.read()
model = model_from_json(model_json, custom_objects=custom_objects)
model.load_weights('%s.h5' % path)
# if verbose: print 'Loaded from %s' % path
return model
json_file = "model.json" # the h5 file should be "model.h5"
model = load_model(json_file) # load the json/h5 pair
model.save('my_saved_model') # this is a directory name to store the saved model