-
Notifications
You must be signed in to change notification settings - Fork 32
/
Visualize.py
156 lines (142 loc) · 5.01 KB
/
Visualize.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
# -*- coding: utf-8 -*-
"""
Created on Wed May 11 17:03:16 2016
@author: Sun
"""
import numpy as np
import matplotlib.pyplot as plt
import cv2
import os
import sys
caffe_root = 'F:/caffe-Microsoft/Build/x64/Release/pycaffe'
#caffe_root = 'F:/caffe-windows/caffe/caffe-windows-master/python'
sys.path.insert(0,caffe_root)
import caffe
#初始化网络
def Init_net(network_proto_path,network_model_path,ImagePath):
caffe.set_mode_cpu()
net = caffe.Net(network_proto_path, network_model_path, caffe.TEST)
transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})
transformer.set_raw_scale('data', 1) # the reference model operates on images in [0,255] range instead of [0,1]
# transformer.set_mean('data', None)
net.blobs['data'].reshape(1,3,128,128)
img = cv2.imread(ImagePath)
plt.imshow(img)
shape0 = img.shape
if shape0[0]!=128 and shape0[1]!=128:
img = cv2.resize(img,(128,128))
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
#plt.imshow(gray)
#gray = gray / 256.0
net.blobs['data'].data[...] = transformer.preprocess('data', gray)
out = net.forward()
return net,transformer
'''
函数:vis_square()
函数功能:一个显示辅助函数
输入参数:data
out----保存图片的路径
'''
def vis_square(data,isLayer='False',out='None'):
#normalize data for display
data = (data - data.min()) / (data.max() - data.min())
#force the number of filters to be square
n = int(np.ceil(np.sqrt(data.shape[0])))
padding = (((0, n ** 2 - data.shape[0]),
(0, 1), (0, 1)) # add some space between filters
+ ((0, 0),) * (data.ndim - 3)) # don't pad the last dimension (if there is one)
data = np.pad(data, padding, mode='constant', constant_values=1) # pad with ones (white)
# tile the filters into an image
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.savefig()
if isLayer == 'True':
cv2.imshow('image',data)
cv2.waitKey(0)
cv2.destroyWindow('image')
else:
cv2.imwrite(out,data*255)
#plt.imshow(data)
#plt.axis('off')
'''
函数:VisualWeights()
函数功能:根据层名字显示相应的权重
输入参数:layer_name----层名字
net----初始化后的网络
out----保存图片的路径
'''
def VisualWeights(net,out):
#显示各个层的shape
# the parameters are a list of [weights, biases]
sub_dir = out + '/' + 'Weights'
if not os.path.exists(sub_dir):
os.mkdir(sub_dir)
for layer_name,param in net.params.iteritems():
out_name = sub_dir + '/' + layer_name + '.jpg'
filters = net.params[layer_name][0].data
vis_square(filters.transpose(0,2,3,1),out_name)
'''
函数:VisualWeight()
函数功能:根据层名字显示相应的权重
输入参数:layer_name----层名字
net----初始化后的网络
'''
def VisualWeight(net,layer_name):
filters = net.params[layer_name][0].data
vis_square(filters.transpose(0,2,3,1),'True')
'''
函数:VisualBlob()
函数功能:可视化Blob
输入参数:net----初始化后的网络
layer_name----层的名字
'''
def VisualBlob(net,layer_name):
feat = net.blobs[layer_name].data[0,]
vis_square(feat,'True')
'''
函数:VisualBlobs()
函数功能:保存所有Blobs
输入参数:net----初始化后的网络
out----保存路径
'''
def VisualBlobs(net,out):
sub_dir = out + '/' + 'Blobs'
if not os.path.exists(sub_dir):
os.mkdir(sub_dir)
for layer_name,blob in net.blobs.iteritems():
out_name = sub_dir + '/' + layer_name + '.jpg'
feat = net.blobs['conv1'].data[0,]
vis_square(feat,out_name)
'''
函数:VisualLayerShape()
函数:显示层的名字和形状
输入参数:net----初始化后的网络
'''
def VisualLayerShape():
#显示各个层的shape
for layer_name1,blob in net.blobs.iteritems():
print layer_name1 + '\t' + str(blob.data.shape)
'''
函数:VisualLayerWeightShape()
函数功能:显示各层权重的形状
输入参数:net----初始化后的网络
'''
def VisualLayerWeightShape(net):
#显示各个层权重的形状
#权重的形状((output_channels, input_channels, filter_height, filter_width))
#param[0]=weights param[1]=biases
for layer_name,param in net.params.iteritems():
print layer_name + '\t' + str(param[0].data.shape),str(param[1].data.shape)
if __name__=='__main__':
network_proto_path = r'F:/Net_train.prototxt'
network_model_path = r'F:/Net_iter_800000.caffemodel'
ImagePath = 'E:/Face_data/FaceImages/0/000000000.bmp'
net,transformer = Init_net(network_proto_path,network_model_path,ImagePath)
#net.params
layer_name = r'fc1'
VisualLayerShape()
#VisualWeights(net,'F:/Weights')
#VisualWeight(net,'conv2a')
#VisualLayerWeightShape(net)
VisualBlob(net,'conv1')
VisualBlobs(net,out='F:/OUT')