环境：
- windows10 64bit
- python: 3.7
- opencv 4.2.0
- tensorflow: 2.1
**目的：**利用opencv中的dnn模块对tensorflow模型进行加载。
opencv的dnn模块有函数dnn.readNetFromTensorflow，根据函数文档可知是调用pb格式的tensorflow模型，这里就入坑了，tensorflow保存的文件格式多种多种：TFLite, frozen graph, SavedModel, serving model, TFHub representation, Keras's .h5 ，tensorflow2.X版本之后推荐使用keras，原版keras默认保存的模型文件是.h5格式的，而tf.keras 模型的save方法默认保存格式是tensorflow的SavedModel（可以通过参数save_format控制），这种方法报道模型也有一个pb文件：
但如果使用dnn模块的相关接口去调用是无法正确读入模型的，经过查阅之后发现需要保存为frozen graph格式，目前网上搜到的大部分关于Keras模型转Frozen grpah的教程所依赖的tensorflow版本都较老，笔者使用的tf版本为最新的2.1版本，最终找到了一个靠谱的方法：
https://leimao.github.io/blog/Save-Load-Inference-From-TF2-Frozen-Graph/
对应的代码在https://github.com/leimao/Frozen_Graph_TensorFlow/blob/master/TensorFlow_v2/train.py
下面笔者提供一份简单代码来演示opencv如何加载tensorflow（keras）模型：
from cv2 import dnn
import cv2
import numpy as np
import matplotlib.pyplot as plt
import os
from keras import backend as K
from keras.models import load_model
#from tensorflow_serving.session_bundle import exporter
from keras.models import model_from_config
from keras.models import Sequential,Model
import tensorflow as tf
from tensorflow.python.tools import freeze_graph
from tensorflow.python.framework.convert_to_constants import convert_variables_to_constants_v2
import os
print(tf.__version__)
print(cv2.__version__)
#%% opencv处理适合模型输入的图片
img_file = r"C:\Users\zhou-\Pictures\cat.jpg"
img_cv2 = cv2.imread(img_file)
print("[INFO]Image shape: ", img_cv2.shape)
# 主要图片尺寸要和模型输入匹配（mobilenet要求输入的尺寸为224*224）
inWidth = 224
inHeight = 224
blob = cv2.dnn.blobFromImage(img_cv2,
scalefactor=1.0 / 255,
size=(inWidth, inHeight),
mean=(0, 0, 0),
swapRB=False,
crop=False)
# blob = np.transpose(blob, (0,2,3,1)) # 适合keras mobilenet网络输入格式
print("[INFO]img shape: ", blob.shape)
#%% 保存keras模型为SaveModel会报错，相关issue见：
# https://github.com/opencv/opencv/issues/16582
model = tf.keras.applications.mobilenet.MobileNet(weights='imagenet')
# model.save('my_model', save_format='tf') # Save model to SavedModel format
# 参考https://github.com/leimao/Frozen_Graph_TensorFlow/blob/master/TensorFlow_v2/train.py
# Save model to SavedModel format
# tf.saved_model.save(model, r"./models")
# Convert Keras model to ConcreteFunction
full_model = tf.function(lambda x: model(x))
full_model = full_model.get_concrete_function(
tf.TensorSpec(model.inputs[0].shape, model.inputs[0].dtype))
# Get frozen ConcreteFunction
frozen_func = convert_variables_to_constants_v2(full_model)
frozen_func.graph.as_graph_def()
layers = [op.name for op in frozen_func.graph.get_operations()]
print("-" * 50)
print("Frozen model layers: ")
for layer in layers:
print(layer)
print("-" * 50)
print("Frozen model inputs: ")
print(frozen_func.inputs)
print("Frozen model outputs: ")
print(frozen_func.outputs)
# Save frozen graph from frozen ConcreteFunction to hard drive
tf.io.write_graph(graph_or_graph_def=frozen_func.graph,
logdir="./frozen_models",
name="frozen_graph.pb",
as_text=False)
net = dnn.readNetFromTensorflow('frozen_models/frozen_graph.pb')
# Run a model
net.setInput(blob)
out = net.forward()
# Get a class with a highest score.
out = out.flatten()
classId = np.argmax(out)
confidence = out[classId]
# Put efficiency information.
t, _ = net.getPerfProfile()
label = 'Inference time: %.2f ms' % (t * 1000.0 / cv2.getTickFrequency())
cv2.putText(img_cv2, label, (0, 15), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0))
# Print predicted class.
def load_imagenet_classes(file_path):
'''
imagenet对应的标签数据如下所示：
0: 'tench, Tinca tinca',
1: 'goldfish, Carassius auratus',
...
'''
classes = []
contents = None
with open(file_path,'r') as f:
contents = f.readlines()
for cnt in contents:
cnt = cnt.strip()
classes.append(cnt.split(':')[1].strip().replace(',',''))
return classes
classes = load_imagenet_classes('imagenet_classes.txt')
label = '%s: %.4f' % (classes[classId] if classes else 'Class #%d' % classId, confidence)
cv2.putText(img_cv2, label, (0, 40), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0))
cv2.imwrite('output-{}.png'.format(img_file.split('\\')[-1][:-4]), img_cv2)