1. VGG-16介绍
vgg是在Very Deep Convolutional Networks for Large-Scale Image Recognition期刊上提出的。模型可以达到92.7%的测试准确度,在ImageNet的前5位。它的数据集包括1400万张图像，1000个类别。
vgg-16是一种深度卷积神经网络模型，16表示其深度，在图像分类等任务中取得了不错的效果。
vgg16 的宏观结构图如下。代码定义在tensorflow的vgg16.py文件 。注意，包括一个预处理层，使用RGB图像在0-255范围内的像素值减去平均值（在整个ImageNet图像训练集计算）。
2. 文件组成
模型权重 - vgg16_weights.npz
TensorFlow模型- vgg16.py
类名（输出模型到类名的映射） - imagenet_classes.py
示例图片输入 - laska.png
我们使用特定的工具转换了原作者在GitHub profile上公开可用的Caffe权重，并做了一些后续处理，以确保模型符合TensorFlow标准。最终实现可用的权重文件vgg16_weights.npz
下载所有的文件到同一文件夹下，然后运行 python vgg16.py
- vgg16.py文件代码：
import tensorflow as tf
import numpy as np
from scipy.misc import imread, imresize
from imagenet_classes import class_names
class vgg16:
def __init__(self, imgs, weights=None, sess=None):
self.imgs = imgs
self.convlayers()
self.fc_layers()
self.probs = tf.nn.softmax(self.fc3l)
if weights is not None and sess is not None:
self.load_weights(weights, sess)
def convlayers(self):
self.parameters = []
# zero-mean input
with tf.name_scope('preprocess') as scope:
mean = tf.constant([123.68, 116.779, 103.939], dtype=tf.float32, shape=[1, 1, 1, 3], name='img_mean')
images = self.imgs-mean
# conv1_1
with tf.name_scope('conv1_1') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 3, 64], dtype=tf.float32,
stddev=1e-1), name='weights')
conv = tf.nn.conv2d(images, kernel, [1, 1, 1, 1], padding='SAME')
biases = tf.Variable(tf.constant(0.0, shape=[64], dtype=tf.float32),
trainable=True, name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv1_1 = tf.nn.relu(out, name=scope)
self.parameters += [kernel, biases]
# conv1_2
with tf.name_scope('conv1_2') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 64, 64], dtype=tf.float32,
stddev=1e-1), name='weights')
conv = tf.nn.conv2d(self.conv1_1, kernel, [1, 1, 1, 1], padding='SAME')
biases = tf.Variable(tf.constant(0.0, shape=[64], dtype=tf.float32),
trainable=True, name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv1_2 = tf.nn.relu(out, name=scope)
self.parameters += [kernel, biases]
# pool1
self.pool1 = tf.nn.max_pool(self.conv1_2,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME',
name='pool1')
# conv2_1
with tf.name_scope('conv2_1') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 64, 128], dtype=tf.float32,
stddev=1e-1), name='weights')
conv = tf.nn.conv2d(self.pool1, kernel, [1, 1, 1, 1], padding='SAME')
biases = tf.Variable(tf.constant(0.0, shape=[128], dtype=tf.float32),
trainable=True, name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv2_1 = tf.nn.relu(out, name=scope)
self.parameters += [kernel, biases]
# conv2_2
with tf.name_scope('conv2_2') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 128, 128], dtype=tf.float32,
stddev=1e-1), name='weights')
conv = tf.nn.conv2d(self.conv2_1, kernel, [1, 1, 1, 1], padding='SAME')
biases = tf.Variable(tf.constant(0.0, shape=[128], dtype=tf.float32),
trainable=True, name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv2_2 = tf.nn.relu(out, name=scope)
self.parameters += [kernel, biases]
# pool2
self.pool2 = tf.nn.max_pool(self.conv2_2,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME',
name='pool2')
# conv3_1
with tf.name_scope('conv3_1') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 128, 256], dtype=tf.float32,
stddev=1e-1), name='weights')
conv = tf.nn.conv2d(self.pool2, kernel, [1, 1, 1, 1], padding='SAME')
biases = tf.Variable(tf.constant(0.0, shape=[256], dtype=tf.float32),
trainable=True, name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv3_1 = tf.nn.relu(out, name=scope)
self.parameters += [kernel, biases]
# conv3_2
with tf.name_scope('conv3_2') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 256, 256], dtype=tf.float32,
stddev=1e-1), name='weights')
conv = tf.nn.conv2d(self.conv3_1, kernel, [1, 1, 1, 1], padding='SAME')
biases = tf.Variable(tf.constant(0.0, shape=[256], dtype=tf.float32),
trainable=True, name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv3_2 = tf.nn.relu(out, name=scope)
self.parameters += [kernel, biases]
# conv3_3
with tf.name_scope('conv3_3') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 256, 256], dtype=tf.float32,
stddev=1e-1), name='weights')
conv = tf.nn.conv2d(self.conv3_2, kernel, [1, 1, 1, 1], padding='SAME')
biases = tf.Variable(tf.constant(0.0, shape=[256], dtype=tf.float32),
trainable=True, name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv3_3 = tf.nn.relu(out, name=scope)
self.parameters += [kernel, biases]
# pool3
self.pool3 = tf.nn.max_pool(self.conv3_3,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME',
name='pool3')
# conv4_1
with tf.name_scope('conv4_1') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 256, 512], dtype=tf.float32,
stddev=1e-1), name='weights')
conv = tf.nn.conv2d(self.pool3, kernel, [1, 1, 1, 1], padding='SAME')
biases = tf.Variable(tf.constant(0.0, shape=[512], dtype=tf.float32),
trainable=True, name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv4_1 = tf.nn.relu(out, name=scope)
self.parameters += [kernel, biases]
# conv4_2
with tf.name_scope('conv4_2') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 512, 512], dtype=tf.float32,
stddev=1e-1), name='weights')
conv = tf.nn.conv2d(self.conv4_1, kernel, [1, 1, 1, 1], padding='SAME')
biases = tf.Variable(tf.constant(0.0, shape=[512], dtype=tf.float32),
trainable=True, name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv4_2 = tf.nn.relu(out, name=scope)
self.parameters += [kernel, biases]
# conv4_3
with tf.name_scope('conv4_3') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 512, 512], dtype=tf.float32,
stddev=1e-1), name='weights')
conv = tf.nn.conv2d(self.conv4_2, kernel, [1, 1, 1, 1], padding='SAME')
biases = tf.Variable(tf.constant(0.0, shape=[512], dtype=tf.float32),
trainable=True, name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv4_3 = tf.nn.relu(out, name=scope)
self.parameters += [kernel, biases]
# pool4
self.pool4 = tf.nn.max_pool(self.conv4_3,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME',
name='pool4')
# conv5_1
with tf.name_scope('conv5_1') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 512, 512], dtype=tf.float32,
stddev=1e-1), name='weights')
conv = tf.nn.conv2d(self.pool4, kernel, [1, 1, 1, 1], padding='SAME')
biases = tf.Variable(tf.constant(0.0, shape=[512], dtype=tf.float32),
trainable=True, name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv5_1 = tf.nn.relu(out, name=scope)
self.parameters += [kernel, biases]
# conv5_2
with tf.name_scope('conv5_2') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 512, 512], dtype=tf.float32,
stddev=1e-1), name='weights')
conv = tf.nn.conv2d(self.conv5_1, kernel, [1, 1, 1, 1], padding='SAME')
biases = tf.Variable(tf.constant(0.0, shape=[512], dtype=tf.float32),
trainable=True, name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv5_2 = tf.nn.relu(out, name=scope)
self.parameters += [kernel, biases]
# conv5_3
with tf.name_scope('conv5_3') as scope:
kernel = tf.Variable(tf.truncated_normal([3, 3, 512, 512], dtype=tf.float32,
stddev=1e-1), name='weights')
conv = tf.nn.conv2d(self.conv5_2, kernel, [1, 1, 1, 1], padding='SAME')
biases = tf.Variable(tf.constant(0.0, shape=[512], dtype=tf.float32),
trainable=True, name='biases')
out = tf.nn.bias_add(conv, biases)
self.conv5_3 = tf.nn.relu(out, name=scope)
self.parameters += [kernel, biases]
# pool5
self.pool5 = tf.nn.max_pool(self.conv5_3,
ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1],
padding='SAME',
name='pool4')
def fc_layers(self):
# fc1
with tf.name_scope('fc1') as scope:
shape = int(np.prod(self.pool5.get_shape()[1:]))
fc1w = tf.Variable(tf.truncated_normal([shape, 4096],
dtype=tf.float32,
stddev=1e-1), name='weights')
fc1b = tf.Variable(tf.constant(1.0, shape=[4096], dtype=tf.float32),
trainable=True, name='biases')
pool5_flat = tf.reshape(self.pool5, [-1, shape])
fc1l = tf.nn.bias_add(tf.matmul(pool5_flat, fc1w), fc1b)
self.fc1 = tf.nn.relu(fc1l)
self.parameters += [fc1w, fc1b]
# fc2
with tf.name_scope('fc2') as scope:
fc2w = tf.Variable(tf.truncated_normal([4096, 4096],
dtype=tf.float32,
stddev=1e-1), name='weights')
fc2b = tf.Variable(tf.constant(1.0, shape=[4096], dtype=tf.float32),
trainable=True, name='biases')
fc2l = tf.nn.bias_add(tf.matmul(self.fc1, fc2w), fc2b)
self.fc2 = tf.nn.relu(fc2l)
self.parameters += [fc2w, fc2b]
# fc3
with tf.name_scope('fc3') as scope:
fc3w = tf.Variable(tf.truncated_normal([4096, 1000],
dtype=tf.float32,
stddev=1e-1), name='weights')
fc3b = tf.Variable(tf.constant(1.0, shape=[1000], dtype=tf.float32),
trainable=True, name='biases')
self.fc3l = tf.nn.bias_add(tf.matmul(self.fc2, fc3w), fc3b)
self.parameters += [fc3w, fc3b]
def load_weights(self, weight_file, sess):
weights = np.load(weight_file)
keys = sorted(weights.keys())
for i, k in enumerate(keys):
print i, k, np.shape(weights[k])
sess.run(self.parameters[i].assign(weights[k]))
if __name__ == '__main__':
sess = tf.Session()
imgs = tf.placeholder(tf.float32, [None, 224, 224, 3])
vgg = vgg16(imgs, 'vgg16_weights.npz', sess)
img1 = imread('laska.png', mode='RGB')
img1 = imresize(img1, (224, 224))
prob = sess.run(vgg.probs, feed_dict={vgg.imgs: [img1]})[0]
preds = (np.argsort(prob)[::-1])[0:5]
for p in preds:
#print class_names[p], prob[p]
print("class_name {}: step {}".format(class_names[p], prob[p]))
运行，测试
测试1：
输入图片为laska.png
运行结果：
2018-03-23 11:04:38.311802: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
2018-03-23 11:04:38.311873: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.
0 conv1_1_W (3, 3, 3, 64)
1 conv1_1_b (64,)
2 conv1_2_W (3, 3, 64, 64)
3 conv1_2_b (64,)
4 conv2_1_W (3, 3, 64, 128)
5 conv2_1_b (128,)
6 conv2_2_W (3, 3, 128, 128)
7 conv2_2_b (128,)
8 conv3_1_W (3, 3, 128, 256)
9 conv3_1_b (256,)
10 conv3_2_W (3, 3, 256, 256)
11 conv3_2_b (256,)
12 conv3_3_W (3, 3, 256, 256)
13 conv3_3_b (256,)
14 conv4_1_W (3, 3, 256, 512)
15 conv4_1_b (512,)
16 conv4_2_W (3, 3, 512, 512)
17 conv4_2_b (512,)
18 conv4_3_W (3, 3, 512, 512)
19 conv4_3_b (512,)
20 conv5_1_W (3, 3, 512, 512)
21 conv5_1_b (512,)
22 conv5_2_W (3, 3, 512, 512)
23 conv5_2_b (512,)
24 conv5_3_W (3, 3, 512, 512)
25 conv5_3_b (512,)
26 fc6_W (25088, 4096)
27 fc6_b (4096,)
28 fc7_W (4096, 4096)
29 fc7_b (4096,)
30 fc8_W (4096, 1000)
31 fc8_b (1000,)
class_name **weasel**: step 0.693385839462
class_name polecat, fitch, foulmart, foumart, Mustela putorius: step 0.175387635827
class_name mink: step 0.12208583951
class_name black-footed ferret, ferret, Mustela nigripes: step 0.00887066219002
class_name otter: step 0.000121083263366分类结果为weasel
测试2：
输入图片为多场景
运行结果为：
2018-03-23 11:15:22.718228: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
2018-03-23 11:15:22.718297: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.
0 conv1_1_W (3, 3, 3, 64)
1 conv1_1_b (64,)
2 conv1_2_W (3, 3, 64, 64)
3 conv1_2_b (64,)
4 conv2_1_W (3, 3, 64, 128)
5 conv2_1_b (128,)
6 conv2_2_W (3, 3, 128, 128)
7 conv2_2_b (128,)
8 conv3_1_W (3, 3, 128, 256)
9 conv3_1_b (256,)
10 conv3_2_W (3, 3, 256, 256)
11 conv3_2_b (256,)
12 conv3_3_W (3, 3, 256, 256)
13 conv3_3_b (256,)
14 conv4_1_W (3, 3, 256, 512)
15 conv4_1_b (512,)
16 conv4_2_W (3, 3, 512, 512)
17 conv4_2_b (512,)
18 conv4_3_W (3, 3, 512, 512)
19 conv4_3_b (512,)
20 conv5_1_W (3, 3, 512, 512)
21 conv5_1_b (512,)
22 conv5_2_W (3, 3, 512, 512)
23 conv5_2_b (512,)
24 conv5_3_W (3, 3, 512, 512)
25 conv5_3_b (512,)
26 fc6_W (25088, 4096)
27 fc6_b (4096,)
28 fc7_W (4096, 4096)
29 fc7_b (4096,)
30 fc8_W (4096, 1000)
31 fc8_b (1000,)
class_name alp: step 0.830908000469
class_name church, church building: step 0.0817768126726
class_name castle: step 0.024959910661
class_name valley, vale: step 0.0158758834004
class_name monastery: step 0.0100631769747分类结果把高山，教堂，城堡，山谷，修道院都识别出来了，效果非常不错，虽然各种精度不高，但是类别是齐全的。
测试3：
输入图片为
运行结果为
2018-03-23 11:34:50.490069: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
2018-03-23 11:34:50.490137: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.
0 conv1_1_W (3, 3, 3, 64)
1 conv1_1_b (64,)
2 conv1_2_W (3, 3, 64, 64)
3 conv1_2_b (64,)
4 conv2_1_W (3, 3, 64, 128)
5 conv2_1_b (128,)
6 conv2_2_W (3, 3, 128, 128)
7 conv2_2_b (128,)
8 conv3_1_W (3, 3, 128, 256)
9 conv3_1_b (256,)
10 conv3_2_W (3, 3, 256, 256)
11 conv3_2_b (256,)
12 conv3_3_W (3, 3, 256, 256)
13 conv3_3_b (256,)
14 conv4_1_W (3, 3, 256, 512)
15 conv4_1_b (512,)
16 conv4_2_W (3, 3, 512, 512)
17 conv4_2_b (512,)
18 conv4_3_W (3, 3, 512, 512)
19 conv4_3_b (512,)
20 conv5_1_W (3, 3, 512, 512)
21 conv5_1_b (512,)
22 conv5_2_W (3, 3, 512, 512)
23 conv5_2_b (512,)
24 conv5_3_W (3, 3, 512, 512)
25 conv5_3_b (512,)
26 fc6_W (25088, 4096)
27 fc6_b (4096,)
28 fc7_W (4096, 4096)
29 fc7_b (4096,)
30 fc8_W (4096, 1000)
31 fc8_b (1000,)
class_name cup: step 0.543631911278
class_name coffee mug: step 0.364796578884
class_name pitcher, ewer: step 0.0259610358626
class_name eggnog: step 0.0117611540481
class_name water jug: step 0.00806392729282分类结果为cup
测试4：
输入图片为
运行结果为
2018-03-23 11:37:23.573090: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
2018-03-23 11:37:23.573159: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.
0 conv1_1_W (3, 3, 3, 64)
1 conv1_1_b (64,)
2 conv1_2_W (3, 3, 64, 64)
3 conv1_2_b (64,)
4 conv2_1_W (3, 3, 64, 128)
5 conv2_1_b (128,)
6 conv2_2_W (3, 3, 128, 128)
7 conv2_2_b (128,)
8 conv3_1_W (3, 3, 128, 256)
9 conv3_1_b (256,)
10 conv3_2_W (3, 3, 256, 256)
11 conv3_2_b (256,)
12 conv3_3_W (3, 3, 256, 256)
13 conv3_3_b (256,)
14 conv4_1_W (3, 3, 256, 512)
15 conv4_1_b (512,)
16 conv4_2_W (3, 3, 512, 512)
17 conv4_2_b (512,)
18 conv4_3_W (3, 3, 512, 512)
19 conv4_3_b (512,)
20 conv5_1_W (3, 3, 512, 512)
21 conv5_1_b (512,)
22 conv5_2_W (3, 3, 512, 512)
23 conv5_2_b (512,)
24 conv5_3_W (3, 3, 512, 512)
25 conv5_3_b (512,)
26 fc6_W (25088, 4096)
27 fc6_b (4096,)
28 fc7_W (4096, 4096)
29 fc7_b (4096,)
30 fc8_W (4096, 1000)
31 fc8_b (1000,)
class_name cellular telephone, cellular phone, cellphone, cell, mobile phone: step 0.465327292681
class_name iPod: step 0.10543012619
class_name radio, wireless: step 0.0810257941484
class_name hard disc, hard disk, fixed disk: step 0.0789099931717
class_name modem: step 0.0603163056076分类结果为 cellular telephone
测试5：
输入图片为
运行结果为
2018-03-23 11:40:40.956946: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
2018-03-23 11:40:40.957016: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.
0 conv1_1_W (3, 3, 3, 64)
1 conv1_1_b (64,)
2 conv1_2_W (3, 3, 64, 64)
3 conv1_2_b (64,)
4 conv2_1_W (3, 3, 64, 128)
5 conv2_1_b (128,)
6 conv2_2_W (3, 3, 128, 128)
7 conv2_2_b (128,)
8 conv3_1_W (3, 3, 128, 256)
9 conv3_1_b (256,)
10 conv3_2_W (3, 3, 256, 256)
11 conv3_2_b (256,)
12 conv3_3_W (3, 3, 256, 256)
13 conv3_3_b (256,)
14 conv4_1_W (3, 3, 256, 512)
15 conv4_1_b (512,)
16 conv4_2_W (3, 3, 512, 512)
17 conv4_2_b (512,)
18 conv4_3_W (3, 3, 512, 512)
19 conv4_3_b (512,)
20 conv5_1_W (3, 3, 512, 512)
21 conv5_1_b (512,)
22 conv5_2_W (3, 3, 512, 512)
23 conv5_2_b (512,)
24 conv5_3_W (3, 3, 512, 512)
25 conv5_3_b (512,)
26 fc6_W (25088, 4096)
27 fc6_b (4096,)
28 fc7_W (4096, 4096)
29 fc7_b (4096,)
30 fc8_W (4096, 1000)
31 fc8_b (1000,)
class_name water bottle: step 0.75726544857
class_name pop bottle, soda bottle: step 0.0976340323687
class_name nipple: step 0.0622750669718
class_name water jug: step 0.0233819428831
class_name soap dispenser: step 0.0179366543889分类结果为 water bottle
参考文档