caffe特征可视化

这篇博文对于caffe 网络训练到的特征进行可视化。

参考：  http://nbviewer.jupyter.org/github/BVLC/caffe/blob/master/examples/00-classification.ipynb

           http://www.cnblogs.com/louyihang-loves-baiyan/p/5134671.html

#-*- coding: UTF-8 -*-import numpy as npimport matplotlib.pyplot as pltimport osimport caffeimport sysimport pickleimport cv2caffe_root = '/home/zhuangni/code/caffe-master/'  deployPrototxt =  '/home/zhuangni/code/Multi-Task/experiment/vgg_1/feature/deploy.prototxt'modelFile = '/home/zhuangni/code/Multi-Task/experiment/vgg_1/model/snapshot_iter_200000.caffemodel.h5'meanFile = '/home/zhuangni/code/Multi-Task/data_CelebA_0_1_224/mean.binaryproto'#网络初始化def initilize():    print 'initilize ... '    sys.path.insert(0, caffe_root + 'python')    caffe.set_mode_cpu()    #caffe.set_device(0)    net = caffe.Net(deployPrototxt, modelFile,caffe.TEST)    return net#取出网络中的params和net.blobs的中的数据def getNetDetails(image, net):    proto_data = open(meanFile, "rb").read()      a = caffe.io.caffe_pb2.BlobProto.FromString(proto_data)      mean  = caffe.io.blobproto_to_array(a)[0]      # input preprocessing: 'data' is the name of the input blob == net.inputs[0]    transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})    transformer.set_transpose('data', (2,0,1))    transformer.set_mean('data', mean) # mean pixel    transformer.set_raw_scale('data', 255)      # the reference model operates on images in [0,255] range instead of [0,1]    transformer.set_channel_swap('data', (2,1,0))      # the reference model has channels in BGR order instead of RGB    # set net to batch size of 50    net.blobs['data'].reshape(1,3,224,224)    net.blobs['data'].data[...] = transformer.preprocess('data', caffe.io.load_image(image))    out = net.forward()        #网络提取conv1_1的输出特征    filters = net.params['conv1_1'][0].data    vis_square(filters.transpose(0, 2, 3, 1),'filters.png')    #conv1_1的特征图    feat = net.blobs['conv1_1'].data[0, :36]    vis_square(feat,'feat.png',padval=1)    pool = net.blobs['pool1'].data[0,:36]    vis_square(pool,'pool.png',padval=1)    #全连接层fc14_attr1的输出值以及输出的正值的直方图    feat = net.blobs['fc14_attr1'].data[0]    plt.subplot(2, 1, 1)    plt.plot(feat.flat)    plt.subplot(2, 1, 2)    _ = plt.hist(feat.flat[feat.flat > 0], bins=100)    plt.savefig('fc.png')    plt.show()    #最后一层的输出概率    feat = net.blobs['fc16_attr1'].data[0]    plt.figure(figsize=(15, 3))    plt.plot(feat.flat)    plt.savefig('prob.png')    plt.show()# 此处将卷积图和进行显示，def vis_square(data, fn, padsize=1, padval=0 ):    data -= data.min()    data /= data.max()        #让合成图为方    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))    #合并卷积图到一个图像中        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:])    print data.shape    plt.imshow(data)    plt.axis('off')    plt.savefig(fn)    plt.show()if __name__ == "__main__":    net = initilize()    testimage = '/home/zhuangni/code/Multi-Task/experiment/vgg_1/feature/202599.jpg'    getNetDetails(testimage, net)

实验结果：

1.网络提取conv1_1的输出特征

  conv1_1 的num_output: 64,则生成64个方格小块.

2. conv1_1的特征图

  选取其中的36个得到的特征图。

3. pool1的特征图

4.全连接层fc14_attr1的输出值以及输出的正值的直方图

    fc14_attr1层的num_output: 4096,横坐标为层输出数目，纵坐标为层输出的值。

5. 最后一层的输出概率