根据训练好的cifar10数据的model，从测试图片中选出一张进行测试，并进行网络模型、卷积结果及参数可视化
注意：本文中代码运行在windows+ipython notebook下，已事先配置好caffe的python接口
导入必需的包
1
2
3
4
5
6
7
8
9
|
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import caffe
%matplotlib inline
plt.rcParams['figure.figsize'] = (8, 8)
plt.rcParams['image.interpolation'] = 'nearest'
plt.rcParams['image.cmap'] = 'gray'
|
载入网络模型
1
2
3
4
5
6
|
# 载入模型，并显示各层数据信息
caffe.set_mode_gpu()
net = caffe.Net('examples/cifar10/cifar10_quick.prototxt',
'examples/cifar10/cifar10_quick_iter_5000.caffemodel.h5',
caffe.TEST)
[(k, v.data.shape) for k, v in net.blobs.items()]
|
[('data', (1L, 3L, 32L, 32L)),
('conv1', (1L, 32L, 32L, 32L)),
('pool1', (1L, 32L, 16L, 16L)),
('conv2', (1L, 32L, 16L, 16L)),
('pool2', (1L, 32L, 8L, 8L)),
('conv3', (1L, 64L, 8L, 8L)),
('pool3', (1L, 64L, 4L, 4L)),
('ip1', (1L, 64L)),
('ip2', (1L, 10L)),
('prob', (1L, 10L))]
可视化网络模型
使用GraphViz+Caffe的draw_net.py来可视化网络模型
1
2
3
|
Rem 运行以下命令前必需先安装配置GraphViz
Rem --rankdir参数为网络方向,BT代表图片上网络从底至顶绘出
python ./Build/x64/Release/pycaffe/draw_net.py examples/cifar10/cifar10_quick_train_test.prototxt examples/cifar10/cifar-quick.png --rankdir=BT
|
Drawing net to examples/cifar10/cifar-quick.png
1
2
3
4
|
#显示模型图片
net_im = mpimg.imread('examples/cifar10/cifar-quick.png')
plt.imshow(net_im)
plt.axis('off')
|
(-0.5, 904.5, 2079.5, -0.5)
加载测试图片
1
2
3
4
5
|
#加载测试图片，并显示
im = caffe.io.load_image('examples/cifar10/cat.jpg')
print im.shape
plt.imshow(im)
plt.axis('off')
|
(1200L, 1600L, 3L)
(-0.5, 1599.5, 1199.5, -0.5)
转换均值
1
2
3
4
5
6
7
8
|
# 编写一个函数，将二进制的均值转换为python的均值
def convert_mean(binMean,npyMean):
blob = caffe.proto.caffe_pb2.BlobProto()
bin_mean = open(binMean, 'rb' ).read()
blob.ParseFromString(bin_mean)
arr = np.array( caffe.io.blobproto_to_array(blob) )
npy_mean = arr[0]
np.save(npyMean, npy_mean )
|
1
2
3
4
|
# 调用函数转换均值
binMean='examples/cifar10/mean.binaryproto'
npyMean='examples/cifar10/mean.npy'
convert_mean(binMean,npyMean)
|
将图片载入Blob
1
2
3
4
5
6
7
8
|
#将图片载入blob中,并减去均值
transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})
transformer.set_transpose('data', (2,0,1))
transformer.set_mean('data', np.load(npyMean).mean(1).mean(1))
transformer.set_raw_scale('data', 255)
transformer.set_channel_swap('data', (2,1,0))
net.blobs['data'].data[...] = transformer.preprocess('data',im)
inputData=net.blobs['data'].data
|
1
2
3
4
5
6
7
8
|
#显示减去均值前后的数据
plt.figure()
plt.subplot(1,2,1),plt.title("origin")
plt.imshow(im)
plt.axis('off')
plt.subplot(1,2,2),plt.title("subtract mean")
plt.imshow(transformer.deprocess('data', inputData[0]))
plt.axis('off')
|
(-0.5, 31.5, 31.5, -0.5)
编写用于参数/卷积结果可视化的函数
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
|
# 编写一个函数，用于显示各层数据
def show_data(data, padsize=1, padval=0):
data -= data.min()
data /= data.max()
n = int(np.ceil(np.sqrt(data.shape[0])))
padding = ((0, n ** 2 - data.shape[0]), (0, padsize), (0, padsize)) + ((0, 0),) * (data.ndim - 3)
data = np.pad(data, padding, mode='constant', constant_values=(padval, padval))
data = data.reshape((n, n) + data.shape[1:]).transpose((0, 2, 1, 3) + tuple(range(4, data.ndim + 1)))
data = data.reshape((n * data.shape[1], n * data.shape[3]) + data.shape[4:])
plt.figure()
plt.imshow(data,cmap='gray')
plt.axis('off')
|
可视化各层数据
1
2
3
4
5
6
|
# 运行模型并显示第一个卷积层的输出数据和权值（filter）
net.forward()
print net.blobs['conv1'].data[0].shape
show_data(net.blobs['conv1'].data[0])
print net.params['conv1'][0].data.shape
show_data(net.params['conv1'][0].data.reshape(32*3,5,5))
|
(32L, 32L, 32L)
(32L, 3L, 5L, 5L)
1
2
3
|
# 显示第一次pooling后的输出数据
show_data(net.blobs['pool1'].data[0])
net.blobs['pool1'].data.shape
|
(1L, 32L, 16L, 16L)
1
2
3
4
5
|
# 显示第二次卷积后的输出数据以及相应的权值（filter）
show_data(net.blobs['conv2'].data[0],padval=0.5)
print net.blobs['conv2'].data.shape
show_data(net.params['conv2'][0].data.reshape(32**2,5,5))
print net.params['conv2'][0].data.shape
|
(1L, 32L, 16L, 16L)
(32L, 32L, 5L, 5L)
1
2
3
4
5
|
# 显示第三次卷积后的输出数据以及相应的权值（filter）,取前１024个进行显示
show_data(net.blobs['conv3'].data[0],padval=0.5)
print net.blobs['conv3'].data.shape
show_data(net.params['conv3'][0].data.reshape(64*32,5,5)[:1024])
print net.params['conv3'][0].data.shape
|
(1L, 64L, 8L, 8L)
(64L, 32L, 5L, 5L)
1
2
3
|
# 显示第三次池化后的输出数据
show_data(net.blobs['pool3'].data[0],padval=0.2)
print net.blobs['pool3'].data.shape
|
(1L, 64L, 4L, 4L)
1
2
3
4
|
# 最后一层输入属于某个类的概率
feat = net.blobs['prob'].data[0]
print feat
plt.plot(feat.flat)
|
[ 0.00170287 0.00115923 0.0225699 0.60395384 0.00453733 0.14171894
0.00307363 0.01260873 0.15008588 0.05858969]
[<matplotlib.lines.Line2D at 0x3bd38080>]
与cifar10中的10种类型名称进行对比：
airplane、automobile、bird、cat、deer、dog、frog、horse、ship、truck
根据测试结果，判断为Cat。