crowd counting——利用高斯核函数将坐标点转换为density map

在Crowd counting领域，常用的训练方法是estimate density map，这种数据可利用高斯核函数得来，在MCNN论文中有具体过程；

也可参加这篇博客：https://blog.csdn.net/zxs0222/article/details/107900465

存储density map的文件目前有两种：.h5文件和.cvs文件，而数据集中常见的是.mat文件，其中记录着所有人头的中心坐标点。

利用.mat生成.h5的过程可参见这篇博客：https://blog.csdn.net/zxs0222/article/details/108910388

而有些代码中定义的是.cvs训练文件，因此本篇内容主要记录根据.mat生成.cvs的过程，根据不同的数据集，代码略有不同，本文以ShanghaiTech A/B、UCF_CC_50以及WorldExpo 50为例：

代码在 Matlab 中完成

1、高斯核函数，放在get_density_map_gaussian.m文件中，写好的函数，不用动：

function im_density = get_density_map_gaussian(im,points)


im_density = zeros(size(im)); 
[h,w] = size(im_density);

if(length(points)==0)
    return;
end

if(length(points(:,1))==1)
    x1 = max(1,min(w,round(points(1,1))));
    y1 = max(1,min(h,round(points(1,2))));
    im_density(y1,x1) = 255;
    return;
end
for j = 1:length(points) 	
    f_sz = 15;
    sigma = 4.0;
    H = fspecial('Gaussian',[f_sz, f_sz],sigma);
    x = min(w,max(1,abs(int32(floor(points(j,1)))))); 
    y = min(h,max(1,abs(int32(floor(points(j,2))))));
    if(x > w || y > h)
        continue;
    end
    x1 = x - int32(floor(f_sz/2)); y1 = y - int32(floor(f_sz/2));
    x2 = x + int32(floor(f_sz/2)); y2 = y + int32(floor(f_sz/2));
    dfx1 = 0; dfy1 = 0; dfx2 = 0; dfy2 = 0;
    change_H = false;
    if(x1 < 1)
        dfx1 = abs(x1)+1;
        x1 = 1;
        change_H = true;
    end
    if(y1 < 1)
        dfy1 = abs(y1)+1;
        y1 = 1;
        change_H = true;
    end
    if(x2 > w)
        dfx2 = x2 - w;
        x2 = w;
        change_H = true;
    end
    if(y2 > h)
        dfy2 = y2 - h;
        y2 = h;
        change_H = true;
    end
    x1h = 1+dfx1; y1h = 1+dfy1; x2h = f_sz - dfx2; y2h = f_sz - dfy2;
    if (change_H == true)
        H =  fspecial('Gaussian',[double(y2h-y1h+1), double(x2h-x1h+1)],sigma);
    end
    im_density(y1:y2,x1:x2) = im_density(y1:y2,x1:x2) +  H;
     
end

end

2、ShanghaiTech A/B 数据集：

图片IMG_1.jpg对应着其坐标标注文件：.GT_IMG_1.mat

用于将.mat生成.csv代码：

clc; clear all;
dataset = 'A/B'; % 这里可以更改A或者B
standard_size = [768,1024];    % 将图像全部调整为这个大小

att = 'test';  % 这里需要在生成csv时调整为train或者test

dataset_name = ['shanghaitech_part_' dataset];
path = ['F:/ShanghaiTech_Part_B/part_' dataset '_final/' att '_data/images/'];
output_path = 'F:/ShanghaiTech_Part_B/';    %保存生成后的csv文件
train_path_img = strcat(output_path, dataset_name,'/', att, '/img/');
train_path_den = strcat(output_path, dataset_name,'/', att, '/den/');

gt_path = ['F:/ShanghaiTech_Part_B/part_' dataset '_final/' att '_data/ground_truth/'];

mkdir(output_path)
mkdir(train_path_img);
mkdir(train_path_den);

if (dataset == 'A')
    num_images = 300;  % 这里更好的做法是用代码检测文件夹内文件数量；
else
    num_images = 316;
end

for idx = 1:2
    i = idx;
    if (mod(idx,10)==0)
        fprintf(1,'Processing %3d/%d files\n', idx, num_images);
    end
    load(strcat(gt_path, 'GT_IMG_',num2str(i),'.mat')) ;
    input_img_name = strcat(path,'IMG_',num2str(i),'.jpg');
    im = imread(input_img_name);  
    [h, w, c] = size(im);
    annPoints =  image_info{1}.location;

    rate_h = standard_size(1)/h;
    rate_w = standard_size(2)/w;

    im = imresize(im,[standard_size(1),standard_size(2)]);
    annPoints(:,1) = annPoints(:,1)*double(rate_w);
    annPoints(:,2) = annPoints(:,2)*double(rate_h);
    
    im_density = get_density_map_gaussian(im,annPoints); 
    im_density = im_density(:,:,1);
    
    imwrite(im, [train_path_img num2str(idx) '.jpg']);
    csvwrite([train_path_den num2str(idx) '.csv'], im_density);
end

3、UCF_CC_50数据集，这个数据集的1.jpg图像对应的.mat文件为：1_ann.mat

相应代码：

clc; clear all;
standard_size = [768,1024];  % 调整图像尺寸，全部统一

path ='F:\UCF_CC_50\';
output_path = './data/UCF_CC_50/';
train_path_img = strcat(output_path,'/', 'img/');
train_path_den = strcat(output_path,'/', 'den/');
train_path_seg = strcat(output_path,'/', 'seg/');

mkdir(output_path)
for i_folder = 1:5
    mkdir([train_path_img, num2str(i_folder)]);
    mkdir([train_path_den, num2str(i_folder)]);
    mkdir([train_path_seg, num2str(i_folder)]);
end
cnt = zeros(5,1);
for idx = 1:50
    i = idx;
    if (mod(idx,10)==0)
        fprintf(1,'Processing %3d/%d files\n', idx, 50);
    end
    load(strcat(path, num2str(i),'_ann.mat')) ;
    input_img_name = strcat(path,num2str(i),'.jpg');
    im = imread(input_img_name);  
    [h, w, c] = size(im);
    if (c == 3)
        im = rgb2gray(im);
    end

    rate_h = standard_size(1)/h;
    rate_w = standard_size(2)/w;
    im = imresize(im,[standard_size(1),standard_size(2)]);
    annPoints(:,1) = annPoints(:,1)*double(rate_w);
    annPoints(:,2) = annPoints(:,2)*double(rate_h);
    
    im_density = get_density_map_gaussian(im,annPoints); 
    im_density = im_density(:,:,1);
    

   i_corss = ceil(rand(1)*5);
    while cnt(i_corss)>=10
        i_corss = ceil(rand(1)*5);
    end

    cnt(i_corss) = cnt(i_corss) +1;
    
    imwrite(im, [train_path_img num2str(i_corss) '/' num2str(idx) '.jpg']);
    csvwrite([train_path_den num2str(i_corss) '/' num2str(idx) '.csv'], im_density);
end

4、WorldExpo 50 数据集，数据格式为：1.jpg对应着1.mat标注文件

代码：

clc; clear all;

fileFolder=fullfile('F:\World 10\WorldExpo10\train_label');
dirOutput=dir(fullfile(fileFolder,'*'));
fileNames={dirOutput.name}';

standard_size = [768,1024];

dataset_name = ['WorldExpo10'];
original_path = ['F:/World 10/WorldExpo10/'];
output_path = 'F:/World 10/data/';
att = 'test';
train_path_img = strcat(output_path, 'train_frame/');
mkdir(train_path_img);

for ii = 3:105  % 从3开始
  gt_path = ['F:/World 10/WorldExpo10/train_label/' fileNames{ii} '/'];
  
  train_path_den = strcat(output_path, 'train_lable/', fileNames{ii}, '/');
  mkdir(train_path_den);
  
  matFolder=fullfile(gt_path);
  matdirOutput=dir(fullfile(matFolder,'*'));
  matNames={matdirOutput.name}';
  num_images=length(matdirOutput)-1;
  
  for idx = 3:num_images
    i = idx;
    if (mod(idx,10)==0)
        fprintf(1,'Processing %3d/%d files\n', idx, num_images);
    end
    load(strcat(gt_path, matNames{idx}));
    input_img_name = strcat(original_path,'train_frame/',strrep(matNames{idx}, 'mat', 'jpg'));
    im = imread(input_img_name);
    [h, w, c] = size(im);
    annPoints =  point_position;

    rate_h = standard_size(1)/h;
    rate_w = standard_size(2)/w;

    im = imresize(im,[standard_size(1),standard_size(2)]);
    annPoints(:,1) = annPoints(:,1)*double(rate_w);
    annPoints(:,2) = annPoints(:,2)*double(rate_h);
    
    im_density = get_density_map_gaussian(im,annPoints); % 调用1. 中的高斯函数
    im_density = im_density(:,:,1);
    
    imwrite(im, [output_path 'train_frame/' strrep(matNames{idx}, 'mat', 'jpg')]);
    csvwrite([output_path 'train_lable/' fileNames{ii} '/' strrep(matNames{idx}, 'mat', 'csv')], im_density);
  end
end