Keras Tensorflow下训练出来的模型在Visual Studio C++环境中的应用

将Keras训练的.hdf5模型固化为.pb模型
from tensorflow.python.framework import graph_io
from keras.models import load_model
from keras import backend as K
import tensorflow as tf

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.global_variables()).difference(keep_var_names or []))
output_names = output_names or []
output_names += [v.op.name for v in tf.global_variables()]
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

"""----------------------------------配置路径-----------------------------------"""
output_path = "D:/sift"
pb_model_name = "resunet_membrane2.pb"

"""----------------------------------导入keras模型------------------------------"""
K.set_learning_phase(0)
net_model = load_model('D:/sift/resunet_membrane120.hdf500-4')

print('input is :', net_model.input.name)
print('output is:', net_model.output.name)

"""----------------------------------保存为.pb格式------------------------------"""
sess = K.get_session()
frozen_graph = freeze_session(K.get_session(), output_names=[net_model.output.op.name])
graph_io.write_graph(frozen_graph, output_path, "resunet_membrane2.pb", as_text=False)
测试.pb模型在tensorflow下的效果

from keras.models import load_model
import matplotlib.pyplot as plt
import numpy as np
import cv2

# data
img_path = "./0.png"
img = cv2.imread(img_path, 0)

# model
keras_model_path = "./resunet_membrane120.hdf500-4"
ResUNnet = load_model(keras_model_path)

# prepare_and_predict
img = img / 255
img = cv2.resize(img, (512, 512))
X = np.reshape(img, (1, 512, 512, 1))
y = ResUNnet.predict(X)
out = np.reshape(y, (512, 512)) * 255
out = np.array(out, dtype="u1")
print("img.shape: %s, out.shape: %s" % (img.shape, out.shape))

# show
plt.subplot(1, 2, 1); plt.imshow(img, cmap="gray")
plt.subplot(1, 2, 2); plt.imshow(out, cmap="gray")
plt.suptitle("keras-hdf5"); plt.show()

"""------------------------tensorflow-pb------------------------"""
import tensorflow as tf

# model

tf_model_path = "./new_tensor_model.pb"

# graph_and_predict
with tf.Graph().as_default():
output_graph_def = tf.GraphDef()

# open the *.pb model
with open(tf_model_path, "rb") as fr:
output_graph_def.ParseFromString(fr.read())
tf.import_graph_def(output_graph_def, name="")

# run the forward in the session
with tf.Session() as sess:
sess.run(tf.global_variables_initializer()) # init
inp_tensor = sess.graph.get_tensor_by_name("input_1:0")#这是对应模型输入张量的名字，每个人的定义不一样，如果你不知道，可以去网上找代码输出所有结点的名字，一般第一个就是
out_tensor = sess.graph.get_tensor_by_name("conv2d_30/Sigmoid:0")#这是输出张量的名字，一般是输出的所有结点的最后一个名字
npy_tensor = sess.run(out_tensor, feed_dict={inp_tensor: X}) # X in line 18.
# postprocessing
npy = np.reshape(npy_tensor, (512, 512)) * 255
npy = np.array(npy, dtype="u1")

# show
plt.subplot(1, 3, 1); plt.imshow(img, cmap="gray"), plt.title("IMG")
plt.subplot(1, 3, 2); plt.imshow(out, cmap="gray"), plt.title("keras")
plt.subplot(1, 3, 3); plt.imshow(npy, cmap="gray"), plt.title("TF")
plt.show()

搭建VS的tensorflow平台

CMake 3.14.1

Swigwin3.0.12

VS2015