-
Notifications
You must be signed in to change notification settings - Fork 0
/
evaluateDukeMTMC_with_SCT.m
212 lines (175 loc) · 8.08 KB
/
evaluateDukeMTMC_with_SCT.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
function result = evaluateDukeMTMC_with_SCT(resMat, iou_threshold, world, testSet)
ROI = getROIs();
if strcmp(testSet,'easy')
load('gt/testData.mat');
gtMat = testData;
testInterval = [263504:356648];
elseif strcmp(testSet,'hard')
load('gt/testHardData.mat');
gtMat = testHardData;
testInterval = [227541:263503];
elseif strcmp(testSet,'trainval')
load('gt/trainval.mat');
gtMat = trainData;
testInterval = [49700:227540]; % takes too long
elseif strcmp(testSet,'trainval_mini') % shorter version of trainval
load('gt/trainval.mat');
gtMat = trainData;
testInterval = [127720:187540];
elseif strcmp(testSet,'val') % approx. last 25 min. of trainval
load('gt/trainval.mat');
gtMat = trainData;
testInterval = [139611:227540];
else
fprintf('Unknown test set %s\n',testSet);
return;
end
% Filter rows by frame interval
startTimes = [5543, 3607, 27244, 31182, 1, 22402, 18968, 46766];
%startTimes = [1, 1, 1, 1, 1, 1, 1, 1];
for cam = 1:8
gtMat(gtMat(:,1) == cam & ~ismember(gtMat(:,3) + startTimes(cam) - 1, testInterval),:) = [];
resMat(resMat(:,1) == cam & ~ismember(resMat(:,3) + startTimes(cam) - 1, testInterval),:) = [];
end
% Filter rows by feet position within ROI
feetpos = [ resMat(:,4) + 0.5*resMat(:,6), resMat(:,5) + resMat(:,7)];
keep = false(size(resMat,1),1);
for cam = 1:8
camFilter = resMat(:,1) == cam;
keep(camFilter & inpolygon(feetpos(:,1),feetpos(:,2), ROI{cam}(:,1),ROI{cam}(:,2))) = true;
end
resMat = resMat(keep,:);
%{
% Filter the result tracklets with duration smaller than a threshold to speed up evaluation
fprintf('filtering tracklets to speed up evaluation...\n');
tracklet_th = 10;
% Mat is in format [cam, ID, frame, x, y, w, h]
resID = unique(resMat(:, 2));
length(resID)
len_stats = zeros(1, length(resID));
for i = 1:length(resID)
ID = resID(i);
match_idx = (resMat(:,2) == ID);
len = sum(match_idx(:));
if len < tracklet_th
resMat(match_idx, :) = [];
end
len_stats(1, i) = len;
if mod(i, floor(length(resID)/10)) == 0
fprintf('%d%% \n', floor(100.*i/length(resID)));
end
end
resID = unique(resMat(:, 2));
length(resID)
%}
% Single-Cam
for camera = 1:8
fprintf('Processing camera %d...\n',camera);
resMatSingle = resMat(resMat(:,1)==camera, 2:7);
gtMatSingle = gtMat(gtMat(:,1)==camera, 2:7);
clust_measures = CLUSTmeasures(resMatSingle, gtMatSingle, iou_threshold, world);
measures = IDmeasures(resMatSingle, gtMatSingle, iou_threshold, world);
result{camera}.CLUSTmeasures = clust_measures;
result{camera}.IDmeasures = measures;
result{camera}.description = sprintf('Cam_%d',camera);
result{camera}.allMets = evaluateTracking(result{camera}.description, gtMatSingle, resMatSingle);
result{camera}.allMets.mets2d.m = [clust_measures.clustF1, clust_measures.clustP, clust_measures.clustR,...
measures.IDF1, measures.IDP, measures.IDR, ...
result{camera}.allMets.mets2d.m];
%fprintf('clustP = %f, clustR= %f', clust_measures.clustP, clust_measures.clustR);
end
fprintf('\n');
% Multi-Cam
% Convert data format to:
% ID, frame, left, top, width, height, worldX, worldY
SHIFT_CONSTANT = 100000000;
gtMatMulti = gtMat(:,2:7);
resMatMulti = resMat(:,2:7);
gtMatMulti(:,2) = gtMat(:,3) + gtMat(:,1)*SHIFT_CONSTANT; % frame + cam*1000000 for frame uniqueness
resMatMulti(:,2) = resMat(:,3) + resMat(:,1)*SHIFT_CONSTANT;
result{10}.description = 'Multi-cam';
result{10}.IDmeasures = IDmeasures(resMatMulti, gtMatMulti, iou_threshold, world);
%result{10}.CLUSTmeasures = CLUSTmeasures(resMatMulti, gtMatMulti, iou_threshold, world);
% Constructing clust_mat from clust_mat from each camera is faster than reconstructing
result{10}.CLUSTmeasures = CLUSTmeasures_aggregate(result, 8);
% AllCameraSingle (MC Upper bound)
gtMatSingleAll = gtMat(:,2:7);
resMatSingleAll = resMat(:,2:7);
gtMatSingleAll(:,1) = gtMatSingleAll(:,1) + gtMat(:,1)*SHIFT_CONSTANT; % ID + cam*1000000 for ID uniqueness
resMatSingleAll(:,1) = resMatSingleAll(:,1) + resMat(:,1)*SHIFT_CONSTANT;
for cam = 1:8 % frame uniqueness
gtMatSingleAll(gtMat(:,1)==cam,2) = gtMatSingleAll(gtMat(:,1)==cam,2) + (cam-1) * numel(testInterval);
resMatSingleAll(resMat(:,1)==cam,2) = resMatSingleAll(resMat(:,1)==cam,2) + (cam-1) * numel(testInterval);
end
result{9}.description = 'Single-all';
if false
measures = IDmeasures(resMatSingleAll, gtMatSingleAll, iou_threshold, world);
result{9}.IDmeasurs = measures;
result{9}.allMets = evaluateTracking(result{9}.description, gtMatSingleAll, resMatSingleAll);
else
% It is faster to aggregate scores from all cameras than to re-evaluate
MT = 0; PT = 0; ML = 0; FRA = 0;
falsepositives = 0; missed = 0; idswitches = 0;
Fgt = 0; iousum = 0; Ngt = 0; sumg = 0;
Nc = 0;
numGT = 0; numPRED = 0; IDTP = 0; IDFP = 0; IDFN = 0;
CLUSTTP = 0; CLUSTFP = 0; CLUSTFN = 0; CLUSTTN = 0;
for cam = 1:8
numGT = numGT + result{cam}.IDmeasures.numGT;
numPRED = numPRED + result{cam}.IDmeasures.numPRED;
IDTP = IDTP + result{cam}.IDmeasures.IDTP;
IDFN = IDFN + result{cam}.IDmeasures.IDFN;
IDFP = IDFP + result{cam}.IDmeasures.IDFP;
CLUSTTP = CLUSTTP + result{cam}.CLUSTmeasures.TP;
CLUSTFP = CLUSTFP + result{cam}.CLUSTmeasures.FP;
CLUSTFN = CLUSTFN + result{cam}.CLUSTmeasures.FN;
CLUSTTN = CLUSTTN + result{cam}.CLUSTmeasures.TN;
MT = MT + result{cam}.allMets.mets2d.additionalInfo.MT;
PT = PT + result{cam}.allMets.mets2d.additionalInfo.PT;
ML = ML + result{cam}.allMets.mets2d.additionalInfo.ML;
FRA = FRA + result{cam}.allMets.mets2d.additionalInfo.FRA;
Fgt = Fgt + result{cam}.allMets.mets2d.additionalInfo.Fgt;
Ngt = Ngt + result{cam}.allMets.mets2d.additionalInfo.Ngt;
Nc = Nc + sum(result{cam}.allMets.mets2d.additionalInfo.c);
sumg = sumg + sum(result{cam}.allMets.mets2d.additionalInfo.g);
falsepositives = falsepositives + sum(result{cam}.allMets.mets2d.additionalInfo.fp);
missed = missed + sum(result{cam}.allMets.mets2d.additionalInfo.m);
idswitches = idswitches + sum(result{cam}.allMets.mets2d.additionalInfo.mme);
ious = result{cam}.allMets.mets2d.additionalInfo.ious;
td = result{cam}.allMets.mets2d.additionalInfo.td;
iousum = iousum + sum(ious(ious>=td & ious<Inf));
end
CLUSTPrecision = CLUSTTP / (CLUSTTP + CLUSTFP);
CLUSTRecall = CLUSTTP / (CLUSTTP + CLUSTFN);
CLUSTF1 = 2 * CLUSTPrecision * CLUSTRecall /(CLUSTPrecision + CLUSTRecall);
IDPrecision = IDTP / (IDTP + IDFP);
IDRecall = IDTP / (IDTP + IDFN);
IDF1 = 2*IDTP/(numGT + numPRED);
clust_measures.clustP = CLUSTPrecision * 100;
clust_measures.clustR = CLUSTRecall * 100;
clust_measures.clustF1 = CLUSTF1 * 100;
result{9}.CLUSTmeasures = clust_measures;
measures.IDP = IDPrecision * 100;
measures.IDR = IDRecall * 100;
measures.IDF1 = IDF1 * 100;
measures.numGT = numGT;
measures.numPRED = numPRED;
measures.IDTP = IDTP;
measures.IDFP = IDFP;
measures.IDFN = IDFN;
result{9}.IDmeasures = measures;
FAR = falsepositives / Fgt;
MOTP=iousum/Nc * 100; % avg ol
MOTAL=(1-(missed+falsepositives+log10(idswitches+1))/sumg)*100;
MOTA=(1-(missed+falsepositives+idswitches)/sumg)*100;
recall=Nc/sumg*100;
precision=Nc/(falsepositives+Nc)*100;
metrics=[recall, precision, FAR, Ngt, MT, PT, ML, falsepositives, missed, idswitches, FRA, MOTA, MOTP, MOTAL];
result{9}.allMets.mets2d.m = metrics;
%metrics=[recall, precision, FAR, Ngt, MT, PT, ML, falsepositives, missed, idswitches, FRA, MOTA, MOTP, MOTAL];
%result{9}.allMets.mets2d.m = metrics;
result{9}.allMets.mets2d.m = [clust_measures.clustF1, clust_measures.clustP, clust_measures.clustR,...
measures.IDF1, measures.IDP, measures.IDR, ...
result{9}.allMets.mets2d.m];
end
%result{9}.allMets.mets2d.m = [measures.IDF1, measures.IDP, measures.IDR, result{9}.allMets.mets2d.m];