原文地址：http://blog.csdn.NET/u014510375/article/details/51704447

最近读到一篇paper非常有意思，他们把caffe里训练好的模型的参数导出来了，然后…弄到了torch里。所以，今天就来看看怎么导出参数吧。 
为了简单，这次我选的是LeNet

import numpy as np
import scipy.io as sio
import caffedef load():# Load the netcaffe.set_mode_cpu()# You may need to train this caffemodel first# There should be script to help you do the trainingnet = caffe.Net(root + 'lenet.prototxt', root + 'lenet_iter_10000.caffemodel',\caffe.TEST)conv1_w = net.params['conv1'][0].dataconv1_b = net.params['conv1'][1].dataconv2_w = net.params['conv2'][0].dataconv2_b = net.params['conv2'][1].dataip1_w = net.params['ip1'][0].dataip1_b = net.params['ip1'][1].dataip2_w = net.params['ip2'][0].dataip2_b = net.params['ip2'][1].datasio.savemat('conv1_w', {'conv1_w':conv1_w})sio.savemat('conv1_b', {'conv1_b':conv1_b})sio.savemat('conv2_w', {'conv2_w':conv2_w})sio.savemat('conv2_b', {'conv2_b':conv2_b})sio.savemat('ip1_w', {'ip1_w':ip1_w})sio.savemat('ip1_b', {'ip1_b':ip1_b})sio.savemat('ip2_w', {'ip2_w':ip2_w})sio.savemat('ip2_b', {'ip2_b':ip2_b})if __name__ == "__main__":# You will need to change this pathroot = '/Users/yuliangzou/caffe-rc3/examples/mnist/'load()print 'Caffemodel loaded and written to .mat files successfully!'

• 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

• 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

从代码里可以看得很清楚啦，首先导入模型，然后利用net.params就可以获取参数了，另外你也可以利用net.data导出数据进行可视化。当然，在导出参数之前…你必须要跑过一遍，不然你没有这个caffemodel… 
最后…要说一下我最近无聊的时候在github上开了个Naive-CNN的项目，就是….把Caffe里的模型参数导出来，用Matlab或者Python写一遍。目前只做了LeNet。欢迎大家也来玩: 
https://github.com/Yuliang-Zou/Naive-CNN