-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathapply_badacost_detector_to_FDDB.m
302 lines (264 loc) · 10.2 KB
/
apply_badacost_detector_to_FDDB.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
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
function apply_badacost_detector_to_FDDB(dataDir, dataOutputDir, SHRINKAGE, RESAMPLING)
% dataDir - Path of the directory with the prepared face databases data.
% dataOutputDir - Path to store trained detector and detection results in.
if (nargin < 4)
RESAMPLING = 1;
end
if (nargin < 3)
SHRINKAGE = 0.1;
end
mkdir(dataOutputDir);
exp_name='AFLW';
% Size of the search window
MIN_HEIGHT = 40;
BEST_ASPECT_RATIO = 1;
BEST_PADDING_RATIO = 1/8;
% Regularisation
%RESAMPLING = 1;
%SHRINKAGE = 0.1;
% FRAC_FTRS = 1/16;
% Cascade calibration
% USE_CALIBRATION = 1
% CALIBRATION_THR_FRACTION = 1;
% % For testing
OVERLAPING_TP = 0.5;
% Our parameters
imgTestDir = fullfile(dataDir, 'fddb_pdollar_format/images');
lbsTestDir = fullfile(dataDir, 'fddb_pdollar_format/annotations');
pLoad={'lbls',{'Face'},'ilbls',{'DontCare'}};
pLoad = {pLoad{:} 'hRng',[round(MIN_HEIGHT*0.8) inf]};
pLoad = {pLoad{:} 'format', 0};
num_classes = 5+1; % 5 orientations + background
DETECTOR_FILE_PREFIX = [exp_name ...
sprintf('_SHRINKAGE_%f_RESAMPLING_%f_ASPECT_RATIO_%f', ...
SHRINKAGE, ...
RESAMPLING, BEST_ASPECT_RATIO) '_'];
%--------------------------------------------------------------------------
% Now, load the already trained BAdaCost based detector
%--------------------------------------------------------------------------
detectorFile = fullfile(dataOutputDir, [DETECTOR_FILE_PREFIX 'Detector.mat']);
%--------------------------------------------------------------------------
% Load the trained detector
%--------------------------------------------------------------------------
detector = load(detectorFile);
detector = detector.detector;
detectorBAK = detector;
% detector.opts.pNms.type = 'max';
% detector.opts.pNms.overlap = 0.5;
% detector.opts.pNms.ovrDnm = 'min';
detector.opts.pPyramid.nOctUp=1; % Needed in FDDB
detectorPrefix_FDDB = ['FDDB' ...
sprintf('_SHRINKAGE_%f_RESAMPLING_%f_ASPECT_RATIO_%f', ...
detector.opts.pBoost.shrinkage, ...
detector.opts.pBoost.resampling, BEST_ASPECT_RATIO) '_'];
detectorFile_FDDB = [detectorPrefix_FDDB 'Detector.mat'];
save(fullfile(dataOutputDir, detectorFile_FDDB), 'detector');
detector = detectorBAK;
%--------------------------------------------------------------------------
% test detector and plot roc (see acfTest)
%--------------------------------------------------------------------------
pLoad2=pLoad;
if iscell(pLoad)
index = find(strcmp(pLoad2, 'format'));
if ~isempty(index)
pLoad2{index+1} = 0; % Format 0 is all data
end
elseif isstruct(pLoad2)
if isfield(pLoad2, 'format')
pLoad2.format = 0; % Format 0 is all data.
end
end
pLoadTest = {pLoad2{:}};
[miss,roc,gt,dt]=acfTestBadacost('name',detectorPrefix_FDDB,...
'imgDir',imgTestDir,...
'gtDir',lbsTestDir,...
'pLoad',pLoadTest,...
'show',1, ...
'thr', OVERLAPING_TP, ...
'numClasses', num_classes, ...
'savePath', dataOutputDir); % Overlaping threshold for a BoundingBox as TP
% From now on we use FDDB dataset to test ...
exp_name = 'FDDB'
save(fullfile(dataOutputDir, [exp_name '_TEST_RESULTS.mat']), 'miss', 'roc', 'gt', 'dt');
h = figure;
ref = 10.^(-2:.25:0);
lims = [3.1e-3 1e1 .05 1];
color = {'r', 'g', 'b', 'k', 'm', 'c', 'y'};
lineSt = {'-', ':', '--', '.', '-', ':', '--'};
%[fp,tp,score,miss_test] = bbGt('compRoc',gt,dt,1,ref);
[fp,tp,~,miss_test] = bbGt('compRoc',gt,dt,1,ref);
[hs,~,~] =plotRoc([fp tp],'logx',1,'logy',0, 'xLbl', 'fppi',...
'lims', lims, 'color', color{1}, 'lineSt', lineSt{1}, 'smooth', 1, 'fpTarget', ref);
legend_string = sprintf('asp.ratio=%2.2f, pad.ratio=%2.2f, recall (at 1FFPI)=%.2f%%', ...
BEST_ASPECT_RATIO, BEST_PADDING_RATIO, ...
miss_test(end)*100);
legend(hs, legend_string, 'Location', 'Best');
hold off;
saveas(gcf, fullfile(dataOutputDir, [exp_name '_Roc.eps']), 'epsc');
saveas(gcf, fullfile(dataOutputDir, [exp_name '_Roc.png']), 'png');
h = figure;
ref = 10.^(-2:.25:0);
lims = [3.1e-3 1e1 .05 1];
color = {'r', 'g', 'b', 'k', 'm', 'c', 'y'};
lineSt = {'-', ':', '--', '.', '-', ':', '--'};
%[fp,tp,score,miss_test] = bbGt('compRoc',gt,dt,1,ref);
[fp,tp,~,miss_test] = bbGt('compRoc',gt,dt,1,ref);
[hs,~,~] = plotRoc([fp tp],'logx',1,'logy',0, 'xLbl', 'fppi',...
'lims', lims, 'color', color{1}, 'lineSt', lineSt{1}, 'smooth', 1, 'fpTarget', ref);
legend_string = sprintf('asp.ratio=%2.2f, pad.ratio=%2.2f, recall (at 0.1 FFPI)=%.2f%%', ...
BEST_ASPECT_RATIO, BEST_PADDING_RATIO, ...
miss_test(5)*100);
legend(hs, legend_string, 'Location', 'Best');
hold off;
saveas(gcf, fullfile(dataOutputDir, [exp_name '_Roc2.eps']), 'epsc');
saveas(gcf, fullfile(dataOutputDir, [exp_name '_Roc2.png']), 'png');
%--------------------------------------------------------------------------
% Plot results over images.
%--------------------------------------------------------------------------
figure;
IMG_RESULTS_PATH = fullfile(dataOutputDir, 'IMG_RESULTS_FDDB');
mkdir(IMG_RESULTS_PATH);
LABELS_RESULTS_PATH = fullfile(dataOutputDir, 'LABELS_RESULTS_FDDB');
mkdir(LABELS_RESULTS_PATH);
showResOpts ={'evShow',1,'gtShow',1, 'dtShow',1, 'isMulticlass', 1};
imgNms = bbGt('getFiles',{imgTestDir});
% Compute per class average detection bb -> ground thruth bb transform.
% IT IS WRONLY IMPLEMENTED: It should take into account only the
% correct detections.
gt_all = cell2mat(gt(:));
dt_all = cell2mat(dt(:));
gt_all(gt_all(:,5) ~= 1,:) = [];
dt_all(dt_all(:,6) ~= 1,:) = [];
dx = zeros(num_classes-1,1);
dy = zeros(num_classes-1,1);
dw = ones(num_classes-1,1);
dh = ones(num_classes-1,1);
for c=1:num_classes-1
ind_c = find((dt_all(:, 7) == c+1));
gt_c = gt_all(ind_c,:);
dt_c = dt_all(ind_c,:);
for it=1:1
x_dt = dt_c(:,1)-1; % + dt_c(:,3)/2;
y_dt = dt_c(:,2)-1; % + dt_c(:,4)/2;
x_gt = gt_c(:,1)-1; % + gt_c(:,3)/2;
y_gt = gt_c(:,2)-1; % + gt_c(:,4)/2;
% All scale changes.
dw_c = gt_c(:,3)./dt_c(:,3);
dh_c = gt_c(:,4)./dt_c(:,4);
% All traslations scaled
dx_c = (x_gt - x_dt)./dt_c(:,3);
dy_c = (y_gt - y_dt)./dt_c(:,4);
% update the data
delta_dxc = median(dx_c);
delta_dyc = median(dy_c);
dx(c) = dx(c) + delta_dxc;
dy(c) = dy(c) + delta_dyc;
delta_dwc = median(dw_c);
delta_dhc = median(dh_c);
dw(c) = dw(c) * delta_dwc;
dh(c) = dh(c) * delta_dhc;
dt_c(:,1) = dt_c(:,1) + dx(c)*dt_c(:,3); % x update
dt_c(:,2) = dt_c(:,2) + dy(c)*dt_c(:,4); % y update
dt_c(:,3) = dt_c(:,3) .* dw(c); % w update
dt_c(:,4) = dt_c(:,4) .* dh(c); % h update
end
end
% Output transforms.
if (~exist('NICE_VISUALISATION', 'var'))
NICE_VISUALISATION = false;
end
if (~exist('NICE_VISUALISATION_SCORE_THRESHOLD', 'var'))
NICE_VISUALISATION = false;
end
if (~NICE_VISUALISATION)
fid = fopen(fullfile(dataOutputDir,'FDDB_COMPATIBLE_Dets.txt'), 'w');
fid_ellip = fopen(fullfile(dataOutputDir,'FDDB_COMPATIBLE_ELLIPSE_Dets.txt'), 'w');
fid_rect2 = fopen(fullfile(dataOutputDir,'FDDB_COMPATIBLE_RECT2_Dets.txt'), 'w');
end
for i=1:length(imgNms)
file_name = strsplit(imgNms{i}, '/');
file_name = file_name{end};
[pathstr,name,ext]=fileparts(file_name);
I = imread(fullfile(imgTestDir, file_name));
dt_i = dt{i};
gt_i = gt{i};
dt_i(:,6) = ones(size(dt_i, 1), 1);
dt_i(:,7) = dt_i(:,7)-ones(size(dt_i, 1), 1);
if NICE_VISUALISATION
% Show results with nice visualization (removed score < NICE_VISUALIZATION_SCORE_THRESHOLD detections)
showResOpts ={'evShow',0,'gtShow',0, 'dtShow',1, 'isMulticlass', 1, 'dtLs', '--', 'lw', 2, 'cols', 'kry'};
dt_i_nice = dt_i(dt_i(:,5)>=NICE_VISUALISATION_SCORE_THRESHOLD, :);
[hs,hImg] = bbGt('showRes', I, gt_i, dt_i_nice, showResOpts); % multiClass = 1
% Ellipses
for j=1:size(dt_i_nice,1)
lbl = dt_i_nice(j,7);
x1 = dt_i_nice(j,1)-1 + dx(lbl)*dt_i_nice(j,3);
y1 = dt_i_nice(j,2)-1 + dy(lbl)*dt_i_nice(j,4);
w = dt_i_nice(j,3)*dw(lbl);
h = dt_i_nice(j,4)*dh(lbl);
a = h/2; % + h*0.2;
b = w/2; % * 1.1;
xc = x1 + w/2;
yc = y1 + h/2; % - h*0.2;
hold on;
plotEllipse(yc+1, xc+1, b, a, 0.0 ,'g',100, 3);
end ;
hold off;
saveas(gcf, fullfile(IMG_RESULTS_PATH, ['NICE_VISUALISATION_' file_name]), 'png');
else
% Show full results and comparison with ground thruth
showResOpts ={'evShow',1,'gtShow',1, 'dtShow',1, 'isMulticlass', 1};
[hs,hImg] = bbGt('showRes', I, gt_i, dt_i, showResOpts); % multiClass = 1
pos = strfind(file_name, '_');
file_name_orig = fullfile(strrep(file_name(1:pos(end-1)-1), '_', '/'), file_name(pos(end-1)+1:end-4))
% Rectangles/squares
fprintf(fid, '%s\n', file_name_orig);
fprintf(fid, '%d\n', size(dt_i,1));
for j=1:size(dt_i,1)
x1 = dt_i(j,1)-1;
y1 = dt_i(j,2)-1;
w = dt_i(j,3);
h = dt_i(j,4);
fprintf(fid, '%f %f %f %f %f\n', x1, y1, w, h, dt_i(j,5));
end;
% Ellipses
fprintf(fid_ellip, '%s\n', file_name_orig);
fprintf(fid_ellip, '%d\n', size(dt_i,1));
for j=1:size(dt_i,1)
lbl = dt_i(j,7);
x1 = dt_i(j,1)-1 + dx(lbl)*dt_i(j,3);
y1 = dt_i(j,2)-1 + dy(lbl)*dt_i(j,4);
w = dt_i(j,3)*dw(lbl);
h = dt_i(j,4)*dh(lbl);
a = h/2; % + h*0.2;
b = w/2; % * 1.1;
xc = x1 + w/2;
yc = y1 + h/2; % - h*0.2;
fprintf(fid_ellip, '%f %f %f %f %f %f\n', a, b, deg2rad(90), xc, yc, dt_i(j,5));
hold on;
plotEllipse(yc+1, xc+1, b, a, 0.0 ,'b',100, 2);
end ;
hold off;
% Rect2
fprintf(fid_rect2, '%s\n', file_name_orig);
fprintf(fid_rect2, '%d\n', size(dt_i,1));
for j=1:size(dt_i,1)
lbl = dt_i(j,7);
x1 = dt_i(j,1)-1 + dx(lbl)*dt_i(j,3);
y1 = dt_i(j,2)-1 + dy(lbl)*dt_i(j,4);
w = dt_i(j,3)*dw(lbl);
h = dt_i(j,4)*dh(lbl);
hold on;
fprintf(fid_rect2, '%f %f %f %f %f\n', x1, y1, w, h, dt_i(j,5));
plot([x1+1 x1+w-2 x1+w-2 x1 x1], [y1+1 y1+1 y1+h-2 y1+h-2 y1+1], 'm');
end;
hold off;
%pause;
saveas(gcf, fullfile(IMG_RESULTS_PATH, [file_name '.png']), 'png');
end
end
if (~NICE_VISUALISATION)
fclose(fid);
fclose(fid_ellip);
fclose(fid_rect2);
end