Convert mpii annotations from .mat to .json format

mpii_mat2json.m

function mpii_convert_json(  )
% convert mpii annotations .mat file to .json


%% load annotation file
fprintf('Load annotations... ')
data = load('/media/HDD2/Datasets/Human_Pose/mpii/mpii_human_pose_v1_u12_2/mpii_human_pose_v1_u12_1.mat');
fprintf('Done.\n')
%% open file
fprintf('Open file mpii_human_pose_annotations.json\n')
path = '/media/HDD2/Datasets/Human_Pose/mpii/mpii_human_pose_v1_u12_2/';
fileID = fopen(strcat(path, 'mpii_human_pose_annotations.json'),'w');

%% cycle all data and save to file (.json)
fprintf(fileID, '"{');

% annolist
fprintf(fileID, '\\"annolist\\":[');
for i=1:1:size(data.RELEASE.annolist,2)
    fprintf(fileID, '{');
    % image
    fprintf(fileID, '\\"image\\": {');
    fprintf(fileID, '\\"name\\": \\"%s\\"', data.RELEASE.annolist(1,i).image.name);
    fprintf(fileID, '},');
    
    
    % annorect
    fprintf(fileID, '\\"annorect\\": [');
    for j=1:1:size(data.RELEASE.annolist(1,i).annorect,2)
        fprintf(fileID, '{');
        
        flag_comma = 0;
        % x1
        if isfield(data.RELEASE.annolist(1,i).annorect(j),'x1')
             fprintf(fileID, '\\"x1\\": %d', data.RELEASE.annolist(1,i).annorect(j).x1);
             flag_comma = 1;
        end

        % y1
        if isfield(data.RELEASE.annolist(1,i).annorect(j),'y1')
            if flag_comma,  fprintf(fileID, ','); end
            flag_comma = 1;
            fprintf(fileID, '\\"y1\\": %d', data.RELEASE.annolist(1,i).annorect(j).y1);             
        end
        
        % x2
        if isfield(data.RELEASE.annolist(1,i).annorect(j),'x2')
            if flag_comma,  fprintf(fileID, ','); end
            flag_comma = 1;
            fprintf(fileID, '\\"x2\\": %d', data.RELEASE.annolist(1,i).annorect(j).x2);
        end
        
        % y2
        if isfield(data.RELEASE.annolist(1,i).annorect(j),'y2')
            if flag_comma,  fprintf(fileID, ','); end
            flag_comma = 1;
            fprintf(fileID, '\\"y2\\": %d', data.RELEASE.annolist(1,i).annorect(j).y2);
        end
        
        % annopoints
        if isfield(data.RELEASE.annolist(1,i).annorect(j),'annopoints')
            if flag_comma,  fprintf(fileID, ','); end
            flag_comma = 1;
            if isempty(data.RELEASE.annolist(1,i).annorect(j).annopoints)
                fprintf(fileID, '\\"annopoints\\": []');
            else
                %disp(i)
                fprintf(fileID, '\\"annopoints\\": {\\"point\\": [');
                for d=1:1:size(data.RELEASE.annolist(1,i).annorect(j).annopoints.point,2)
                    fprintf(fileID, '{');
                    fprintf(fileID, '\\"x\\": %d, ', data.RELEASE.annolist(1,i).annorect(j).annopoints.point(1,d).x);
                    fprintf(fileID, '\\"y\\": %d, ', data.RELEASE.annolist(1,i).annorect(j).annopoints.point(1,d).y);
                    fprintf(fileID, '\\"id\\": %d ', data.RELEASE.annolist(1,i).annorect(j).annopoints.point(1,d).id);
                    if isfield(data.RELEASE.annolist(1,i).annorect(j).annopoints.point(1,d),'is_visible')
                        fprintf(fileID, ',');
                        if isempty(data.RELEASE.annolist(1,i).annorect(j).annopoints.point(1,d).is_visible)
                            fprintf(fileID, '\\"is_visible\\": []');
                        else
                            fprintf(fileID, '\\"is_visible\\": %d', data.RELEASE.annolist(1,i).annorect(j).annopoints.point(1,d).is_visible);
                        end
                    end
                    fprintf(fileID, '}');

                    if d < size(data.RELEASE.annolist(1,i).annorect(j).annopoints.point,2)
                        fprintf(fileID, ',');
                    end   
                end
                fprintf(fileID, ']}');      
            end
        end
        
        % scale
        if isfield(data.RELEASE.annolist(1,i).annorect(j),'scale')
            if flag_comma,  fprintf(fileID, ','); end
            flag_comma = 1;
            if isempty(data.RELEASE.annolist(1,i).annorect(j).scale)
                fprintf(fileID, '\\"scale\\": []');
            else
                fprintf(fileID, '\\"scale\\": %d', data.RELEASE.annolist(1,i).annorect(j).scale);
            end
        end
        
        % objpos
        if isfield(data.RELEASE.annolist(1,i).annorect(j),'objpos')
            if flag_comma,  fprintf(fileID, ','); end
            flag_comma = 1;
            if isempty(data.RELEASE.annolist(1,i).annorect(j).objpos)
                fprintf(fileID, '\\"objpos\\": []');
            else
                fprintf(fileID, '\\"objpos\\": {');
                fprintf(fileID, '\\"x\\": %d, ', data.RELEASE.annolist(1,i).annorect(j).objpos.x);
                fprintf(fileID, '"y\\": %d', data.RELEASE.annolist(1,i).annorect(j).objpos.y);
                fprintf(fileID, '}');
            end
        end
        
        fprintf(fileID, '}');
        
        if j < size(data.RELEASE.annolist(1,i).annorect,2)
            fprintf(fileID, ',');
        end        
    end
    fprintf(fileID, '],');
    
    
    % frame_sec
    if isempty(data.RELEASE.annolist(1,i).frame_sec)
        fprintf(fileID, '\\"frame_sec\\": [], ');
    else
        fprintf(fileID, '\\"frame_sec\\": %d, ', data.RELEASE.annolist(1,i).frame_sec);
    end

    % vididx
    if isempty(data.RELEASE.annolist(1,i).vididx)
        fprintf(fileID, '\\"vididx\\": [] ');
    else
        fprintf(fileID, '\\"vididx\\": %d ', data.RELEASE.annolist(1,i).vididx);
    end
    
    fprintf(fileID, '}');
    
     if i<size(data.RELEASE.annolist,2) 
        fprintf(fileID, ',');
    end
end
fprintf(fileID, '],');


% img_train
fprintf(fileID, '\\"img_train\\":[');
for i=1:1:size(data.RELEASE.img_train,2)
    fprintf(fileID, '%d', data.RELEASE.img_train(1,i));
    
    if i<size(data.RELEASE.img_train,2) 
        fprintf(fileID, ',');
    end
end
fprintf(fileID, '],');

% version
fprintf(fileID, '\\"version\\": %s,', data.RELEASE.version);

% single_person
fprintf(fileID, '\\"single_person\\":[');
for i=1:1:size(data.RELEASE.single_person,1)
    if isempty( data.RELEASE.single_person{i,1})
         fprintf(fileID, '[]');
    elseif size( data.RELEASE.single_person{i,1},1) == 1
        fprintf(fileID, '%d', data.RELEASE.single_person{i,1});
    else
        fprintf(fileID, '[');
        for j=1:1:size( data.RELEASE.single_person{i,1},1)
            fprintf(fileID, '%d', data.RELEASE.single_person{i,1}(j));
            if j < size( data.RELEASE.single_person{i,1},1)
                fprintf(fileID, ',');
            end
        end
        fprintf(fileID, ']');
    end
    
    if i<size(data.RELEASE.single_person,1) 
        fprintf(fileID, ',');
    end
end
fprintf(fileID, '],');

% act
fprintf(fileID, '\\"act\\":[');
for i=1:1:size(data.RELEASE.act,1)
    fprintf(fileID, '{');
    % cat_name
    if isempty(data.RELEASE.act(i,1).cat_name)
        fprintf(fileID, '\\"cat_name\\": [], ');
    else
        fprintf(fileID, '\\"cat_name\\": \\"%s\\", ', data.RELEASE.act(i,1).cat_name);
    end
    
    % act_name
    if isempty(data.RELEASE.act(i,1).act_name)
        fprintf(fileID, '\\"act_name\\": [], ');
    else
        fprintf(fileID, '\\"act_name\\": \\"%s\\", ', data.RELEASE.act(i,1).act_name);
    end
    
    % act_id
    fprintf(fileID, '\\"act_id\\": %d', data.RELEASE.act(i,1).act_id);
    
    fprintf(fileID, '}');
    
    if i<size(data.RELEASE.act,1) 
        fprintf(fileID, ',');
    end
end
fprintf(fileID, '],');

% video_list
fprintf(fileID, '\\"video_list\\":[');
for i=1:1:size(data.RELEASE.video_list,2)
    fprintf(fileID, '\\"%s\\"', data.RELEASE.video_list{1,i});
    
    if i<size(data.RELEASE.video_list,2) 
        fprintf(fileID, ',');
    end
end
fprintf(fileID, ']');

%% close file
fprintf(fileID, '}');
fclose(fileID);

fprintf('Script complete.\n')
end

The idea to use empty lists instead of 0 and 1s is that you can do extra analysis with that. In cases where you want only samples with full annotated data in order to train a model, or you want to collect data on only one body joint, you can skip everything that is an empty list. Back when I used this dataset, these were some of the use cases that required me to not fill missing values with numbers (thus avoiding introducing bias into the data).

In essence, the goal of this script is to make you explicitly target what you need from the dataset and make you deal with the corner cases.

Thank you for your reply, sincerely. Regards, Jackie Chou 发送自 Windows 10 版邮件应用 发件人: M. Farrajota 发送时间: 2019年6月8日 16:41 收件人: farrajota 抄送: Jackie Chou; Comment 主题: Re: farrajota/mpii_mat2json.m The idea to use empty lists instead of 0 and 1s is that you can do extra analysis with that. In cases where you want only samples with full annotated data in order to train a model, or you want to collect data on only one body joint, you can skip everything that is an empty list. Back when I used this dataset, these were some of the use cases that required me to not fill missing values with numbers (thus avoiding introducing bias into the data). In essence, the goal of this script is to make you explicitly target what you need from the dataset and make you deal with the corner cases. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

JSONDecodeError: Extra data: im getting this error when Im trying to read the file

JSONDecodeError: Extra data: im getting this error when Im trying to read the file

It's been a while but do you remember finding any solution to this?