tensorlayer/example/tutorial_vgg19.py

#! /usr/bin/python
 # -*- coding: utf-8 -*-
 """
 VGG-19 for ImageNet.
  
 Pre-trained model in this example - VGG19 NPZ and
 trainable examples of VGG16/19 in TensorFlow can be found here:
 https://github.com/machrisaa/tensorflow-vgg
  
 For simplified CNN layer see "Convolutional layer (Simplified)"
 in read the docs website.
  
 """
  
 import os
 import time
 import numpy as np
 import skimage
 import skimage.io
 import skimage.transform
 import tensorflow as tf
 import tensorlayer as tl
 from tensorlayer.layers import *
  
 tf.logging.set_verbosity(tf.logging.DEBUG)
 tl.logging.set_verbosity(tl.logging.DEBUG)
  
 try:
 from tensorlayer.models.imagenet_classes import *
 except Exception as e:
 raise Exception(
 "{} / download the file from: https://github.com/tensorlayer/tensorlayer/tree/master/example/data".format(e)
 )
  
 VGG_MEAN = [103.939, 116.779, 123.68]
  
 MODEL_DIR = "models"
 MODEL_NAME = "vgg19.npy"
 MODEL_PATH = os.path.join(MODEL_DIR, MODEL_NAME)
  
  
 def load_image(path):
 # load image
 img = skimage.io.imread(path)
 img = img / 255.0
 if ((0 <= img).all() and (img <= 1.0).all()) is False:
 raise Exception("image value should be [0, 1]")
 # print "Original Image Shape: ", img.shape
 # we crop image from center
 short_edge = min(img.shape[:2])
 yy = int((img.shape[0] - short_edge) / 2)
 xx = int((img.shape[1] - short_edge) / 2)
 crop_img = img[yy:yy + short_edge, xx:xx + short_edge]
 # resize to 224, 224
 resized_img = skimage.transform.resize(crop_img, (224, 224), anti_aliasing=False)
 return resized_img
  
  
 def print_prob(prob):
 synset = class_names
 # print prob
 pred = np.argsort(prob)[::-1]
 # Get top1 label
 top1 = synset[pred[0]]
 print("Top1: ", top1, prob[pred[0]])
 # Get top5 label
 top5 = [(synset[pred[i]], prob[pred[i]]) for i in range(5)]
 print("Top5: ", top5)
 return top1
  
  
 def Vgg19(rgb):
 """
 Build the VGG 19 Model
  
 Parameters
 -----------
 rgb : rgb image placeholder [batch, height, width, 3] values scaled [0, 1]
 """
 start_time = time.time()
 print("build model started")
 rgb_scaled = rgb * 255.0
 # Convert RGB to BGR
 red, green, blue = tf.split(rgb_scaled, 3, 3)
  
 if red.get_shape().as_list()[1:] != [224, 224, 1]:
 raise Exception("image size unmatch")
  
 if green.get_shape().as_list()[1:] != [224, 224, 1]:
 raise Exception("image size unmatch")
  
 if blue.get_shape().as_list()[1:] != [224, 224, 1]:
 raise Exception("image size unmatch")
  
 bgr = tf.concat([
 blue - VGG_MEAN[0],
 green - VGG_MEAN[1],
 red - VGG_MEAN[2],
 ], axis=3)
  
 if bgr.get_shape().as_list()[1:] != [224, 224, 3]:
 raise Exception("image size unmatch")
 # input layer
 net_in = InputLayer(bgr, name='input')
 # conv1
 net = Conv2dLayer(net_in, act=tf.nn.relu, shape=[3, 3, 3, 64], strides=[1, 1, 1, 1], padding='SAME', name='conv1_1')
 net = Conv2dLayer(net, act=tf.nn.relu, shape=[3, 3, 64, 64], strides=[1, 1, 1, 1], padding='SAME', name='conv1_2')
 net = PoolLayer(net, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME', pool=tf.nn.max_pool, name='pool1')
 # conv2
 net = Conv2dLayer(net, act=tf.nn.relu, shape=[3, 3, 64, 128], strides=[1, 1, 1, 1], padding='SAME', name='conv2_1')
 net = Conv2dLayer(net, act=tf.nn.relu, shape=[3, 3, 128, 128], strides=[1, 1, 1, 1], padding='SAME', name='conv2_2')
 net = PoolLayer(net, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME', pool=tf.nn.max_pool, name='pool2')
 # conv3
 net = Conv2dLayer(net, act=tf.nn.relu, shape=[3, 3, 128, 256], strides=[1, 1, 1, 1], padding='SAME', name='conv3_1')
 net = Conv2dLayer(net, act=tf.nn.relu, shape=[3, 3, 256, 256], strides=[1, 1, 1, 1], padding='SAME', name='conv3_2')
 net = Conv2dLayer(net, act=tf.nn.relu, shape=[3, 3, 256, 256], strides=[1, 1, 1, 1], padding='SAME', name='conv3_3')
 net = Conv2dLayer(net, act=tf.nn.relu, shape=[3, 3, 256, 256], strides=[1, 1, 1, 1], padding='SAME', name='conv3_4')
 net = PoolLayer(net, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME', pool=tf.nn.max_pool, name='pool3')
 # conv4
 net = Conv2dLayer(net, act=tf.nn.relu, shape=[3, 3, 256, 512], strides=[1, 1, 1, 1], padding='SAME', name='conv4_1')
 net = Conv2dLayer(net, act=tf.nn.relu, shape=[3, 3, 512, 512], strides=[1, 1, 1, 1], padding='SAME', name='conv4_2')
 net = Conv2dLayer(net, act=tf.nn.relu, shape=[3, 3, 512, 512], strides=[1, 1, 1, 1], padding='SAME', name='conv4_3')
 net = Conv2dLayer(net, act=tf.nn.relu, shape=[3, 3, 512, 512], strides=[1, 1, 1, 1], padding='SAME', name='conv4_4')
 net = PoolLayer(net, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME', pool=tf.nn.max_pool, name='pool4')
 # conv5
 net = Conv2dLayer(net, act=tf.nn.relu, shape=[3, 3, 512, 512], strides=[1, 1, 1, 1], padding='SAME', name='conv5_1')
 net = Conv2dLayer(net, act=tf.nn.relu, shape=[3, 3, 512, 512], strides=[1, 1, 1, 1], padding='SAME', name='conv5_2')
 net = Conv2dLayer(net, act=tf.nn.relu, shape=[3, 3, 512, 512], strides=[1, 1, 1, 1], padding='SAME', name='conv5_3')
 net = Conv2dLayer(net, act=tf.nn.relu, shape=[3, 3, 512, 512], strides=[1, 1, 1, 1], padding='SAME', name='conv5_4')
 net = PoolLayer(net, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME', pool=tf.nn.max_pool, name='pool5')
 # fc 6~8
 net = FlattenLayer(net, name='flatten')
 net = DenseLayer(net, n_units=4096, act=tf.nn.relu, name='fc6')
 net = DenseLayer(net, n_units=4096, act=tf.nn.relu, name='fc7')
 net = DenseLayer(net, n_units=1000, act=None, name='fc8')
 print("build model finished: %fs" % (time.time() - start_time))
 return net
  
  
 def Vgg19_simple_api(rgb):
 """
 Build the VGG 19 Model
  
 Parameters
 -----------
 rgb : rgb image placeholder [batch, height, width, 3] values scaled [0, 1]
 """
 start_time = time.time()
 print("build model started")
 rgb_scaled = rgb * 255.0
 # Convert RGB to BGR
 red, green, blue = tf.split(rgb_scaled, 3, 3)
  
 if red.get_shape().as_list()[1:] != [224, 224, 1]:
 raise Exception("image size unmatch")
  
 if green.get_shape().as_list()[1:] != [224, 224, 1]:
 raise Exception("image size unmatch")
  
 if blue.get_shape().as_list()[1:] != [224, 224, 1]:
 raise Exception("image size unmatch")
  
 bgr = tf.concat([
 blue - VGG_MEAN[0],
 green - VGG_MEAN[1],
 red - VGG_MEAN[2],
 ], axis=3)
  
 if bgr.get_shape().as_list()[1:] != [224, 224, 3]:
 raise Exception("image size unmatch")
  
 # input layer
 net_in = InputLayer(bgr, name='input')
 # conv1
 net = Conv2d(net_in, 64, filter_size=(3, 3), strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv1_1')
 net = Conv2d(net, n_filter=64, filter_size=(3, 3), strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv1_2')
 net = MaxPool2d(net, filter_size=(2, 2), strides=(2, 2), padding='SAME', name='pool1')
 # conv2
 net = Conv2d(net, n_filter=128, filter_size=(3, 3), strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv2_1')
 net = Conv2d(net, n_filter=128, filter_size=(3, 3), strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv2_2')
 net = MaxPool2d(net, filter_size=(2, 2), strides=(2, 2), padding='SAME', name='pool2')
 # conv3
 net = Conv2d(net, n_filter=256, filter_size=(3, 3), strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv3_1')
 net = Conv2d(net, n_filter=256, filter_size=(3, 3), strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv3_2')
 net = Conv2d(net, n_filter=256, filter_size=(3, 3), strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv3_3')
 net = Conv2d(net, n_filter=256, filter_size=(3, 3), strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv3_4')
 net = MaxPool2d(net, filter_size=(2, 2), strides=(2, 2), padding='SAME', name='pool3')
 # conv4
 net = Conv2d(net, n_filter=512, filter_size=(3, 3), strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv4_1')
 net = Conv2d(net, n_filter=512, filter_size=(3, 3), strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv4_2')
 net = Conv2d(net, n_filter=512, filter_size=(3, 3), strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv4_3')
 net = Conv2d(net, n_filter=512, filter_size=(3, 3), strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv4_4')
 net = MaxPool2d(net, filter_size=(2, 2), strides=(2, 2), padding='SAME', name='pool4')
 # conv5
 net = Conv2d(net, n_filter=512, filter_size=(3, 3), strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv5_1')
 net = Conv2d(net, n_filter=512, filter_size=(3, 3), strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv5_2')
 net = Conv2d(net, n_filter=512, filter_size=(3, 3), strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv5_3')
 net = Conv2d(net, n_filter=512, filter_size=(3, 3), strides=(1, 1), act=tf.nn.relu, padding='SAME', name='conv5_4')
 net = MaxPool2d(net, filter_size=(2, 2), strides=(2, 2), padding='SAME', name='pool5')
 # fc 6~8
 net = FlattenLayer(net, name='flatten')
 net = DenseLayer(net, n_units=4096, act=tf.nn.relu, name='fc6')
 net = DenseLayer(net, n_units=4096, act=tf.nn.relu, name='fc7')
 net = DenseLayer(net, n_units=1000, act=None, name='fc8')
 print("build model finished: %fs" % (time.time() - start_time))
 return net
  
  
 sess = tf.InteractiveSession()
 x = tf.placeholder("float", [None, 224, 224, 3])
 # net = Vgg19(x)
 net = Vgg19_simple_api(x)
 y = net.outputs
 probs = tf.nn.softmax(y, name="prob")
 tl.layers.initialize_global_variables(sess)
  
 # You need to download the pre-trained model - VGG19 NPY
 if not os.path.isfile(MODEL_PATH):
 print("Please download vgg19.npy from : https://github.com/machrisaa/tensorflow-vgg")
 exit()
  
 npy_file = np.load(MODEL_PATH, encoding='latin1').item()
  
 params = []
 for val in sorted(npy_file.items()):
 W = np.asarray(val[1][0])
 b = np.asarray(val[1][1])
 print(" Loading %s: %s, %s" % (val[0], W.shape, b.shape))
 params.extend([W, b])
  
 print("Restoring model from npy file")
 tl.files.assign_params(sess, params, net)
  
 img1 = load_image("data/tiger.jpeg") # test data in github
 img1 = img1.reshape((1, 224, 224, 3))
 start_time = time.time()
 prob = sess.run(probs, feed_dict={x: img1})
 print("End time : %.5ss" % (time.time() - start_time))
  
 print_prob(prob[0])
  • 0
    点赞
  • 0
    收藏
    觉得还不错? 一键收藏
  • 0
    评论

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值