caffe权值及featureMap可视化

1、权值可视化

主函数 conv1_weights_vis.m,放在caffe根目录，需要matcaffe

clear;clc;close all;addpath('matlab')caffe.set_mode_cpu();fprintf(['Caffe Version = ', caffe.version(), '\n']);net = caffe.Net('models/bvlc_reference_caffenet/deploy.prototxt', 'models/bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel', 'test');fprintf('Load net done. Net layers : ');net.layer_names fprintf('Net blobs : ');net.blob_names% Conv1 Weight Visualizationconv1_layer = net.layer_vec(2);blob1 = conv1_layer.params(1);w = blob1.get_data();fprintf('Conv1 Weight shape: ');size(w)visualize_weights(w, 1);% Conv2 Weight Visualizationconv2_layer = net.layer_vec(6);blob2 = conv2_layer.params(1);w2 = blob2.get_data();fprintf('Conv2 Weight shape: ');size(w2)visualize_weights(w2, 1);% Conv3 Weight Visualizationconv3_layer = net.layer_vec(10);blob3 = conv3_layer.params(1);w3 = blob3.get_data();fprintf('Conv3 Weight shape: ');size(w3)visualize_weights(w3, 1);% Conv4 Weight Visualizationconv4_layer = net.layer_vec(12);blob4 = conv4_layer.params(1);w4 = blob4.get_data();fprintf('Conv4 Weight shape: ');size(w4)visualize_weights(w4, 1);% Conv5 Weight Visualizationconv5_layer = net.layer_vec(14);blob5 = conv5_layer.params(1);w5 = blob5.get_data();fprintf('Conv5 Weight shape: ');size(w5)visualize_weights(w5, 1);

visualize_weights.m

function [] = visualize_weights(w, s)rr=size(w,1);cc=size(w,2);h = max(rr, cc);             % Kernel sizeg = h + s;          % Grid size, larger than Kernel size for better visual effects.% Normalization for gray scalew = w - min(min(min(min(w))));w = w / max(max(max(max(w)))) * 255;w = uint8(w);W = zeros(g * size(w, 3), g * size(w, 4));for u = 1:size(w, 3)    for v = 1:size(w, 4)        W(g * (u - 1) + (1:cc), g * (v -1) + (1:rr)) = w(:,:,u,v)';       % figure,imshow(uint8(w(:,:,u,v)));    endendW = uint8(W);figure;imshow(W);

第一层：

caffenet 的网络结构图如下：

2、featureMap可视化

主函数段：

clear;clc;close all;addpath('matlab')caffe.set_mode_cpu();fprintf(['Caffe Version = ', caffe.version(), '\n']);net = caffe.Net('models/bvlc_reference_caffenet/deploy.prototxt', 'models/bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel', 'test');fprintf('Load net done. Net layers : ');net.layer_names fprintf('Net blobs : ');net.blob_namesim=imread('examples/images/cat.jpg');figure,imshow(im);title('Original Image');d=load('matlab/+caffe/imagenet/ilsvrc_2012_mean.mat');mean_data=d.mean_data;IMAGE_DIM=256;CROPPED_DIM=227;im_data=im(:,:,[3,2,1]); %matlab图像通道是RGB，转换为opencv格式BGRim_data=permute(im_data,[2,1,3]); %matlab内部是列优先存储，转化为opencv格式的行优先存储im_data=single(im_data); %将uint8格式，转化为single类型。im_data=imresize(im_data,[IMAGE_DIM IMAGE_DIM],'bilinear');im_data=im_data-mean_data;im=imresize(im_data,[CROPPED_DIM CROPPED_DIM],'bilinear');km=cat(4,im,im,im,im,im); % 227*227*3*5pm=cat(4,km,km); % 227*227*3*10  因为输入要求为 input_param { shape: { dim: 10 dim: 3 dim: 227 dim: 227 } }，注意顺序反了input_data={pm};scores=net.forward(input_data);scores=scores{1};scores=mean(scores,2);[~,maxlabel]=max(scores);maxlabelfigure;plot(scores);fm_data=net.blob_vec(1);d1=fm_data.get_data();fprintf('Data size=')size(d1)visualize_feature_maps(d1,1);% 卷积层1fm_conv1=net.blob_vec(2);f1=fm_conv1.get_data();fprintf('Feature map conv1 size=')size(f1)visualize_feature_maps(f1,1);% 卷积层2fm_conv2=net.blob_vec(5);f2=fm_conv2.get_data();fprintf('Feature map conv2 size=')size(f2)visualize_feature_maps(f2,1);% 卷积层3fm_conv3=net.blob_vec(8);f3=fm_conv3.get_data();fprintf('Feature map conv3 size=')size(f3)visualize_feature_maps(f3,1);% 卷积层4fm_conv4=net.blob_vec(9);f4=fm_conv4.get_data();fprintf('Feature map conv4 size=')size(f4)visualize_feature_maps(f4,1);% 卷积层5fm_conv5=net.blob_vec(10);f5=fm_conv5.get_data();fprintf('Feature map conv5 size=')size(f5)visualize_feature_maps(f5,1);

function []=visualize_feature_maps(w,s)h=max(size(w,1),size(w,2));g=h+s;c=size(w,3);cv=ceil(sqrt(c));W=zeros(g*cv,g*cv);%%% 缺少最后一个通道10for u=1:cv    for v=1:cv        tw=zeros(h,h);        if(((u-1)*cv+v)<=c)            tw=w(:,:,(u-1)*cv+v,1)';            tw=tw-min(min(tw));            tw=tw/max(max(tw))*255;        end        W(g*(u-1)+(1:h),g*(v-1)+(1:h))=tw;    endendW=uint8(W);figure,imshow(W);