-
Notifications
You must be signed in to change notification settings - Fork 4
/
TLift.m
76 lines (66 loc) · 3.05 KB
/
TLift.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
function out_score = TLift(in_score, gal_cam_id, gal_time, prob_cam_id, prob_time, ...
num_cams, tau, sigma, K, alpha)
%% Function for the Temporal Lifting (TLift) method
% TLift is a model-free temporal cooccurrence based score weighting method proposed in "Interpretable and Generalizable Person Re-Identification with Query-Adaptive
% Convolution and Temporal Lifting".
% Inputs:
% in_score: the similarity score of size [num_gals, num_probs] between the gallery and probe sets.
% gal_cam_id: camera index for samples in the gallery set.
% gal_time: time stamps of samples in the gallery set.
% prob_cam_id: camera index for samples in the probe set.
% prob_time: time stamps of samples in the probe set.
% num_cams: the number of cameras.
% tau: the interval threshold to define nearby persons.
% sigma: the sensitivity parameter of the time difference.
% K: parameter of the top K retrievals used to define the pivot set P.
% alpha: regularizer for the multiplication fusion.
% All the cam_id and time inputs are column vectors, and they are in the same order corresponding to
% rows (gallery) or columns (probe) of the in_score.
% Outputs:
% out_score: the refined score by TLift, with the same size as the in_score.
% Version:
% 1.01, 16-12-2019
% Author:
% Shengcai Liao
% scliao@ieee.org
% Reference:
% Shengcai Liao and Ling Shao, "Interpretable and Generalizable Person Re-Identification with Query-Adaptive
% Convolution and Temporal Lifting." In The European Conference on Computer Vision (ECCV), 23-28 August, 2020.
num_gals = length(gal_time);
num_probs = length(prob_time);
out_score = zeros(num_gals, num_probs);
prob_time_diff = prob_time - prob_time';
cooccur_mask = (abs(prob_time_diff) < tau);
g_sam_index = cell(num_cams, 1);
score = cell(num_cams, 1);
gal_time_diff = cell(num_cams, 1);
for g_cam = 1 : num_cams
g_sam_index{g_cam} = find(gal_cam_id == g_cam);
score{g_cam} = in_score(g_sam_index{g_cam}, :);
frame_id = gal_time(g_sam_index{g_cam});
gal_time_diff{g_cam} = frame_id - frame_id';
end
for p_cam = 1 : num_cams
p_sam_index = find(prob_cam_id == p_cam);
c_mask = cooccur_mask(p_sam_index, p_sam_index);
np = length(p_sam_index);
for g_cam = 1 : num_cams
% No need to evaluate within-camera pairs, but in some public datasets
% they still evaluate negative pairs in the same camera.
% if p_cam == g_cam
% continue;
% end
prob_score = score{g_cam}(:, p_sam_index);
for i = 1 : np
cooccur_index = find(c_mask(:, i));
cooccur_score = prob_score(:, cooccur_index);
[r, index] = sort(cooccur_score(:), 'descend');
thr = r(K);
mask_in_gal = any(cooccur_score >= thr, 2);
dt = gal_time_diff{g_cam}(:, mask_in_gal);
weight = mean(exp(-dt.^2 / sigma^2), 2);
out_score(g_sam_index{g_cam}, p_sam_index(i)) = weight;
end
end
end
out_score = (out_score + alpha) .* in_score;