/
demo_window.m
123 lines (99 loc) · 4.13 KB
/
demo_window.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
clear all
%%
addpath('ZhuRamananDetector','optimisations','utils','comparison');
% YOU MUST set this to the base directory of the Basel Face Model
BFMbasedir = '\\QCRI-PRECISION-\Documents\ajay\CNN\Face\SOA\3dmm_cnn\3DMM_model\';
% Load morphable model
load(strcat(BFMbasedir,'01_MorphableModel.mat'));
% Important to use double precision for use in optimisers later
shapeEV = double(shapeEV);
shapePC = double(shapePC);
shapeMU = double(shapeMU);
% We need edge-vertex and edge-face lists to speed up finding occluding boundaries
% Either: 1. Load precomputed edge structure for Basel Face Model
load('BFMedgestruct.mat');
% Or: 2. Compute lists for new model:
%TR = triangulation(tl,ones(k,1),ones(k,1),ones(k,1));
%Ev = TR.edges;
%clear TR;
%Ef = meshFaceEdges( tl,Ev );
%save('edgestruct.mat','Ev','Ef');
%% ADJUSTABLE PARAMETERS
% Number of model dimensions to use
ndims = 60;
% Prior weight for initial landmark fitting
w_initialprior=0.7;
% Number of iterations for iterative closest edge fitting
icefniter=7;
options.Ef = Ef;
options.Ev = Ev;
% w1 = weight for edges
% w2 = weight for landmarks
% w3 = 1-w1-w2 = prior weight
options.w1 = 0.45;
options.w2 = 0.15;
%% Setup basic parameters
testdir='\\QCRI-PRECISION-/Documents/ajay/CNN/Face/dataset/mean_img_rotate/';
for img_id = 149:-1:100
if(~exist(fullfile(testdir,sprintf('mean_%03d_1_15.png',img_id))))
continue;
end
im = imread(fullfile(testdir,sprintf('mean_%03d_1_15.png',img_id)));
edgeim = edge(rgb2gray(im),'canny',0.15);
ZRtimestart = tic;
bs = LandmarkDetector(im);
if(isempty(bs))
continue;
end
ZRtime = toc(ZRtimestart);
disp(['Time for landmark detection: ' num2str(ZRtime) ' seconds']);
[x_landmarks,landmarks]=ZR2BFM( bs,im );
%% Initialise using only landmarks
disp('Fitting to landmarks only...');
[b,R,t,s] = FitSingleSOP( x_landmarks,shapePC,shapeMU,shapeEV,ndims,landmarks,w_initialprior );
FV.vertices=reshape(shapePC(:,1:ndims)*b+shapeMU,3,size(shapePC,1)/3)';
FV.faces = tl;
%% Initialise using iterative closest edge fitting (ICEF)
disp('Fitting to edges with iterative closest edge fitting...');
[b,R,t,s] = FitEdges(im,x_landmarks,landmarks,shapePC,shapeMU,shapeEV,options.Ef,options.Ev,tl,ndims, w_initialprior, options.w1, options.w2,icefniter);
FV.vertices = reshape(shapePC(:,1:ndims)*b+shapeMU,3,size(shapePC,1)/3)';
%% Run final optimisation of hard edge cost
disp('Optimising non-convex edge cost...');
maxiter = 5;
iter = 0;
diff = 1;
eps = 1e-9;
[r,c]=find(edgeim);
r = size(edgeim,1)+1-r;
while (iter<maxiter) && (diff>eps)
FV.vertices=reshape(shapePC(:,1:ndims)*b+shapeMU,3,size(shapePC,1)/3)';
[ options.occludingVertices ] = occludingBoundaryVertices( FV,options.Ef,options.Ev,R );
X = reshape(shapePC(:,1:ndims)*b+shapeMU,3,size(shapePC(:,1:ndims),1)/3);
% Compute position of projected occluding boundary vertices
x_edge = R*X(:,options.occludingVertices);
x_edge = x_edge(1:2,:);
x_edge(1,:)=s.*(x_edge(1,:)+t(1));
x_edge(2,:)=s.*(x_edge(2,:)+t(2));
% Find edge correspondences
[idx,d] = knnsearch([c r],x_edge');
% Filter edge matches - ignore the worse 5%
sortedd=sort(d);
threshold = sortedd(round(0.95*length(sortedd)));
idx = idx(d<threshold);
options.occludingVertices = options.occludingVertices(d<threshold);
b0 = b;
[ R,t,s,b ] = optimiseHardEdgeCost( b0,x_landmarks,shapeEV,shapeMU,shapePC,R,t,s,r,c,landmarks,options,tl,false );
diff = norm(b0-b);
disp(num2str(diff));
iter = iter+1;
end
% Run optimisation for a final time but without limit on number of
% iterations
[ R,t,s,b ] = optimiseHardEdgeCost( b,x_landmarks,shapeEV,shapeMU,shapePC,R,t,s,r,c,landmarks,options,tl,true );
save(fullfile('\\QCRI-PRECISION-/Documents/ajay/CNN/Face/SOA/3DMM_edges/mean_img_rotate',sprintf('mean_%03d_15.mat',img_id)),'R','t','s','b');
end
disp('Rendering final results...');
FV.vertices=reshape(shapePC(:,1:ndims)*b+shapeMU,3,size(shapePC,1)/3)';
figure; subplot(1,3,1); patch(FV, 'FaceColor', [1 1 1], 'EdgeColor', 'none', 'FaceLighting', 'phong'); light; axis equal; axis off;
subplot(1,3,2); imshow(renderFace(FV,im,R,t,s,false));
subplot(1,3,3); imshow(renderFace(FV,im,R,t,s,true));