ZF/OpenFace
1
0
Fork 0
mirror of https://gitcode.com/gh_mirrors/ope/OpenFace.git synced 2026-08-13 13:07:46 +00:00
Code Issues Packages Projects Releases Wiki Activity

Incorporating the new face detector in some of the matlab scripts, give a warning if MatConvNet not present

Browse Source
This commit is contained in:
Tadas Baltrusaitis
2017-08-08 10:53:20 -04:00
parent f2c0b5b885
commit f18e4264b8
5 changed files with 50 additions and 51 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
matlab_version/demo/face_image_demo.m
View File

@@ -16,6 +16,10 @@ addpath('../CCNF/');
clmParams.multi_modal_types = patches(1).multi_modal_types;
% Dependencies for face detection (MatConvNet), remove if not present
setup_mconvnet;
addpath('../face_detection/mtcnn/');
%%
root_dir = '../../samples/';
images = dir([root_dir, '*.jpg']);
@@ -25,8 +29,11 @@ verbose = true;
for img=1:numel(images)
image_orig = imread([root_dir images(img).name]);
% MTCNN face detector
[bboxs, det_shapes, confidences] = detect_face_mtcnn(image_orig);
% First attempt to use the Matlab one (fastest but not as accurate, if not present use yu et al.)
[bboxs, det_shapes] = detect_faces(image_orig, {'cascade', 'yu'});
% [bboxs, det_shapes] = detect_faces(image_orig, {'cascade', 'yu'});
% Zhu and Ramanan and Yu et al. are slower, but also more accurate
% and can be used when vision toolbox is unavailable
% [bboxs, det_shapes] = detect_faces(image_orig, {'yu', 'zhu'});
@@ -52,28 +59,14 @@ for img=1:numel(images)
hold on;
end
for i=1:size(bboxs,2)
for i=1:size(bboxs,1)
% Convert from the initial detected shape to CLM model parameters,
% if shape is available
bbox = bboxs(:,i);
if(~isempty(det_shapes))
shape = det_shapes(:,:,i);
inds = [1:60,62:64,66:68];
M = pdm.M([inds, inds+68, inds+68*2]);
E = pdm.E;
V = pdm.V([inds, inds+68, inds+68*2],:);
[ a, R, T, ~, params, err, shapeOrtho] = fit_PDM_ortho_proj_to_2D(M, E, V, shape);
g_param = [a; Rot2Euler(R)'; T];
l_param = params;
bbox = bboxs(i,:);
% Use the initial global and local params for clm fitting in the image
[shape,~,~,lhood,lmark_lhood,view_used] = Fitting_from_bb(image, [], bbox, pdm, patches, clmParams, 'gparam', g_param, 'lparam', l_param);
else
[shape,~,~,lhood,lmark_lhood,view_used] = Fitting_from_bb(image, [], bbox, pdm, patches, clmParams);
end
[shape,~,~,lhood,lmark_lhood,view_used] = Fitting_from_bb(image, [], bbox, pdm, patches, clmParams);
% shape correction for matlab format
shape = shape + 1;

55
matlab_version/demo/face_video_demo.m
View File

@@ -33,6 +33,12 @@ od = cd('../face_validation/');
setup;
cd(od);
% Setup the face detector (remove the setup mconvnet if not using
% MatConvNet)
setup_mconvnet;
addpath('../face_detection/mtcnn/');
%%
for v=1:numel(vids)
% load the video
@@ -66,8 +72,9 @@ for v=1:numel(vids)
image_orig = read(vr, i);
if((~det && mod(i,4) == 0) || ~initialised)
[bboxs, det_shapes, confidences] = detect_face_mtcnn(image_orig);
% First attempt to use the Matlab one (fastest but not as accurate, if not present use yu et al.)
[bboxs, det_shapes] = detect_faces(image_orig, {'cascade', 'yu'});
% [bboxs, det_shapes] = detect_faces(image_orig, {'cascade', 'yu'});
% Zhu and Ramanan and Yu et al. are slower, but also more accurate
% and can be used when vision toolbox is unavailable
% [bboxs, det_shapes] = detect_faces(image_orig, {'yu', 'zhu'});
@@ -75,8 +82,8 @@ for v=1:numel(vids)
if(~isempty(bboxs))
% Pick the biggest face for tracking
[~,ind] = max(bboxs(3,:) - bboxs(1,:));
bbox = bboxs(:,ind);
[~,ind] = max(bboxs(:,3) - bboxs(:,1));
bbox = bboxs(ind,:);
% Discard overly small detections
if(bbox(3) - bbox(1) > 40)
@@ -84,39 +91,27 @@ for v=1:numel(vids)
% Either infer the local and global shape parameters
% from the detected landmarks or just using the
% bounding box
if(~isempty(det_shapes))
shape = det_shapes(:,:,ind);
inds = [1:60,62:64,66:68];
M = pdm.M([inds, inds+68, inds+68*2]);
E = pdm.E;
V = pdm.V([inds, inds+68, inds+68*2],:);
[ a, R, T, ~, params, err] = fit_PDM_ortho_proj_to_2D(M, E, V, shape);
g_param_n = [a; Rot2Euler(R)'; T];
l_param_n = params;
else
num_points = numel(pdm.M) / 3;
num_points = numel(pdm.M) / 3;
M = reshape(pdm.M, num_points, 3);
width_model = max(M(:,1)) - min(M(:,1));
height_model = max(M(:,2)) - min(M(:,2));
M = reshape(pdm.M, num_points, 3);
width_model = max(M(:,1)) - min(M(:,1));
height_model = max(M(:,2)) - min(M(:,2));
a = (((bbox(3) - bbox(1)) / width_model) + ((bbox(4) - bbox(2))/ height_model)) / 2;
a = (((bbox(3) - bbox(1)) / width_model) + ((bbox(4) - bbox(2))/ height_model)) / 2;
tx = (bbox(3) + bbox(1))/2;
ty = (bbox(4) + bbox(2))/2;
tx = (bbox(3) + bbox(1))/2;
ty = (bbox(4) + bbox(2))/2;
% correct it so that the bounding box is just around the minimum
% and maximum point in the initialised face
tx = tx - a*(min(M(:,1)) + max(M(:,1)))/2;
ty = ty + a*(min(M(:,2)) + max(M(:,2)))/2;
% correct it so that the bounding box is just around the minimum
% and maximum point in the initialised face
tx = tx - a*(min(M(:,1)) + max(M(:,1)))/2;
ty = ty + a*(min(M(:,2)) + max(M(:,2)))/2;
% visualisation
g_param_n = [a, 0, 0, 0, tx, ty]';
% visualisation
g_param_n = [a, 0, 0, 0, tx, ty]';
l_param_n = zeros(size(pdm.E));
end
l_param_n = zeros(size(pdm.E));
% If tracking has not started trust the detection
if(~initialised)
@@ -186,7 +181,7 @@ for v=1:numel(vids)
end
hold off;
drawnow expose;
pause(0.05);
pause(0.01);
if(record)
frame = getframe;

6
matlab_version/face_detection/detect_faces.m
View File

@@ -4,8 +4,8 @@ function [ bboxes, shapes ] = detect_faces( image, types )
% image - the image to detect the faces on
% type - cell array of the face detectors to use: 'zhu', 'yu', 'cascade'
% OUTPUT:
% bboxes - a set of bounding boxes describing the detected faces 4 x
% num_faces, the format is [min_x; min_y; max_x; max_y];
% bboxes - a set of bounding boxes describing the detected faces num_faces x
% 4, the format is [min_x; min_y; max_x; max_y];
% shapes - if the face detector detects landmarks as well, output them
% n_points x 2 x num_faces
@@ -57,6 +57,6 @@ function [ bboxes, shapes ] = detect_faces( image, types )
if(use_zhu && isempty(bboxes))
[bboxes, shapes] = Detect_tree_based_zhu(image);
end
bboxes = bboxes''
end

4
matlab_version/face_detection/mtcnn/PReLU.m
View File

@@ -9,6 +9,10 @@ function [ out_map ] = PReLU( input_maps, PReLU_params )
% A more readable but slower version
% in_map = input_maps(:,:,i,:);
% in_map(in_map < 0) = in_map(in_map<0) * PReLU_params(i);
% alternative
% out_map(:,:,i,:) = max(input_maps(:,:,i,:),0) + min(input_maps(:,:,i,:),0)*PReLU_params(i);
out_map(:,:,i,:) = input_maps(:,:,i,:) .* (PReLU_params(i) + (1 - PReLU_params(i)) * (input_maps(:,:,i,:) > 0)) ;
end
else

7
matlab_version/face_detection/mtcnn/detect_face_mtcnn.m
View File

@@ -1,5 +1,10 @@
function [total_bboxes, lmarks, confidence] = detect_face_mtcnn(img, min_face_size)
% Check if MatConvNet is installed
if(~exist('vl_nnconv', 'file') == 3)
fprintf('Warning MatConvNet is not installed or not setup, face detection will be quite slow\n');
end
height_orig = size(img,1);
width_orig = size(img,2);
@@ -211,5 +216,7 @@ new_txs = widths * -0.0075 + txs;
new_tys = heights * 0.2459 + tys;
total_bboxes = [new_txs, new_tys, new_txs + new_widths, new_tys + new_heights];
total_bboxes = double(total_bboxes);
lmarks = double(lmarks);
end