import cv2
import tensorflow as tf
from tensorflow.python.framework import graph_utildef freeze_graph(input_checkpoint ,output_graph):''':param input_checkpoint::param output_graph: PB模型保存路径:return:'''with tf.name_scope('input'):input_data = tf.placeholder(dtype=tf.float32, shape=[1, 256, 512, 3], name='input_data') #定义输入节点net = lanenet.LaneNet(phase='test', net_flag='vgg')ret = net.inference(input_tensor=input_data, name='lanenet_model') # 网络模型结构output = tf.identity(ret, name='output_label') # 定义输出节点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init = tf.global_variables_initializer()sess.run(init)print("load model file success")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_defoutput_node_names=[output.op.name] )# 如果有多个输出节点，以逗号隔开print("持久化成功")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以上代码将ckpy文件转为pb文件
def inference():with tf.gfile.FastGFile('./seglanenet.pb', 'rb') as model_file:graph = tf.Graph().as_default()graph_def = tf.GraphDef()graph_def.ParseFromString(model_file.read())tf.import_graph_def(graph_def,name="")return_elements = tf.import_graph_def(graph_def, return_elements=["input/input_data:0", "output_label:0"]) # 获取对应模型的输入输出节点with tf.Session() as sess:label = sess.run(return_elements[1], feed_dict={return_elements[0]:images})return label以上代码是进行pb文件的模型推理过程。