TensorFlow实现打印每一层的输出_Python

TensorFlow实现打印每一层的输出

2020-04-03 19:27Kluiverthoo Python

今天小编就为大家分享一篇TensorFlow实现打印每一层的输出，具有很好的参考价值，希望对大家有所帮助。一起跟随小编过来看看吧

在test.py中可以通过如下代码直接生成带weight的pb文件，也可以通过tf官方的freeze_graph.py将ckpt转为pb文件。

				?

									constant_graph = graph_util.convert_variables_to_constants(sess, sess.graph_def,['net_loss/inference/encode/conv_output/conv_output'])

									with tf.gfile.FastGFile('net_model.pb', mode='wb') as f:

									  f.write(constant_graph.SerializeToString())

tf1.0中通过带weight的pb文件与get_tensor_by_name函数可以获取每一层的输出

				?

									import os

									import os.path as ops

									import argparse

									import time

									import math

									import tensorflow as tf

									import glob

									import numpy as np

									import matplotlib.pyplot as plt

									import cv2

									os.environ["CUDA_VISIBLE_DEVICES"] = "-1"

									gragh_path = './model.pb'

									image_path = './lvds1901.JPG'

									inputtensorname = 'input_tensor:0'

									tensorname = 'loss/inference/encode/resize_images/ResizeBilinear'

									filepath='./net_output.txt'

									HEIGHT=256

									WIDTH=256

									VGG_MEAN = [103.939, 116.779, 123.68]

									with tf.Graph().as_default():

									  graph_def = tf.GraphDef()

									  with tf.gfile.GFile(gragh_path, 'rb') as fid:

									    serialized_graph = fid.read()

									    graph_def.ParseFromString(serialized_graph)

									    tf.import_graph_def(graph_def, name='')

									    image = cv2.imread(image_path)

									    image = cv2.resize(image, (WIDTH, HEIGHT), interpolation=cv2.INTER_CUBIC)

									    image_np = np.array(image)

									    image_np = image_np - VGG_MEAN

									    image_np_expanded = np.expand_dims(image_np, axis=0)

									    with tf.Session() as sess:

									      ops = tf.get_default_graph().get_operations()

									      tensor_name = tensorname + ':0'

									      tensor_dict = tf.get_default_graph().get_tensor_by_name(tensor_name)

									      image_tensor = tf.get_default_graph().get_tensor_by_name(inputtensorname)

									      output = sess.run(tensor_dict, feed_dict={image_tensor: image_np_expanded})

									      ftxt = open(filepath,'w')

									      transform = output.transpose(0, 3, 1, 2)

									      transform = transform.flatten()

									      weight_count = 0

									      for i in transform:

									        if weight_count % 10 == 0 and weight_count != 0:

									          ftxt.write('\n')

									        ftxt.write(str(i) + ',')

									        weight_count += 1

									      ftxt.close()