demoDMHS.m

function [pose3D, pose2D, bodyPartLabeling] = demoDMHS(img_path, output_name, multiplier_3D, multiplier_sem)

% demoDMHS - apply Deep Multitask Human Sensing (DMHS) CNN model on a single image
% [pose3D, pose2D, bodyPartLabeling] = demoDMHS(img_path, output_name, multiplier_3D, multiplier_sem)
%
% OUTPUT :
% pose3D           - cell array with 6 elements (one for each stage of the
%                    DMHS) each containing the corresponding 3D pose
%                    prediction (17x3)
% pose2D           - cell array with 6 elements (one for each stage of the
%                    DMHS) each containing the corresponding 2D pose
%                    prediction (14x2)
% bodyPartLabeling - cell array with 6 elements (one for each stage of the
%                    DMHS) each containing the corresponding body part
%                    labeling mask
%
% INPUT  :
% img_path         - path to image used for testing
%                  - it should contain a single person inside a bounding box
% output_name      - output name for mat file with results corresponding to
%                    image to img_path
%                  - the mat file will be saved by default in ./data/results/
% multiplier_3D    - image scales used by the 3D pose estimation task of
%                    DMHS network
% multiplier_sem   - image scales used by the body part labeling task of
%                    DMHS network

if(nargin == 0)
    img_path = './data/images/im1020.jpg';
%     img_path = './data/images/im1037.jpg';
%     img_path = './data/images/im1054.jpg';
%     img_path = './data/images/im2673.png';
%     img_path = './data/images/im2788.png';
end
if(nargin < 2)
    output_name = 'results_im1020';
%     output_name = 'results_im1037';
%     output_name = 'results_im1054';
%     output_name = 'results_im2673';
%     output_name = 'results_im2788';
end
if(nargin < 3)
    multiplier_sem = 0.4:0.1:1;
end
if(nargin < 4)
    multiplier_3D = 0.7:0.1:1;
end;

img = imread(img_path);
% Recommended size for the X axis of img in order to obtain the best
% network results with default multiplier_sem (0.4:0.1:1) and multiplier_3D
% (0.7:0.1:1). For different image sizes, please revalidate multiplier_sem
% and multiplier_3D.
avgDimX = 386;
resizeFactor = avgDimX / size(img, 1);
img = imresize(img, resizeFactor);

% switch displayModel to 1 for visualizing the results
displayMode = 1;

addpath('./code/');

% use_gpu = 1 for GPU usage (please note that for all 6 stages it requires
%                            11 GB of GPU memory)
% use_gpu = 0 for CPU usage
use_gpu = 1;
% default model_id for DMHS
model_id = 1;
param = config(use_gpu, 0, model_id);
model = param.model(param.modelID);
net = caffe.Net(model.deployFile, model.caffemodel, 'test');

rectangle = [1 1 size(img, 2) size(img, 1)];
% First output parameter of applyModelPreloadedNet_DMHS returns the cell
% arrays containing the processed outputs of each of the network's stages.
% Second output parameter of applyModelPreloadedNet_DMHS returns the raw
% output of the network.
[predictions, network_output] = applyModelPreloadedNet_DMHS(net, img, param, rectangle, multiplier_3D, multiplier_sem);
pose2D = predictions{1};
bodyPartLabeling = predictions{2};
pose3D = predictions{3};
if (~exist('./data/results/', 'dir'))
    mkdir('./data/results/');
end;
save(['./data/results/' output_name '.mat'], 'pose3D', 'pose2D', 'bodyPartLabeling', 'img', 'network_output');

if (displayMode == 1)
    figure;
    % image used for testing
    imshow(img);
    title('Test Image');
    figure;
    % Estimated 3D poses corresponding to each stage (1 - 6)
    set(gcf, 'Position', get(0,'Screensize'));
    for i = 1 : 6
        subplot(2, 3, i);
        plotSkel3D(pose3D{i}, 'r');
        title(sprintf('Pose 3D prediction - stage %d', i));
    end
    figure;
    % Body part labeling prediction corresponding to each stage (1 - 6)
    set(gcf, 'Position', get(0,'Screensize'));
    for i = 1 : 6
        subplot(2, 3, i);
        imagesc(bodyPartLabeling{i});
        axis image;
        axis off;
        title(sprintf('Body labeling prediction - stage %d', i));
    end
    figure;
    % Estimated 2D poses corresponding to each stage (1 - 6)
    set(gcf, 'Position', get(0,'Screensize'));
    facealpha = 0.6; % for limb transparency
    truncate = zeros(1,model.np);
    for i = 1:6
        pred = pose2D{i};
        subplot(2,3,i);
        imshow(img);
        hold on;
        bodyHeight = max(pred(:,2)) - min(pred(:,2));
        plotSkel2D(model, facealpha, pred, truncate, bodyHeight/30);
        plot(pred(:,1), pred(:,2), 'k.', 'MarkerSize', bodyHeight/32);
        title(sprintf('Pose 2D prediction - stage %d', i));
    end;
end;