Tracktor algorithm

[young-yangb]

Hi,

I have read your tracktor algorithm in your paper, and I'm confused about two parts:

(1) line 13 in Algorithm 2: Tracktor algorithm (public detections)
B ← NMS(B, S, λactive);
for k, Tk ∈ Tactive do
if k /∈ B then
T← T+ {Tk};
Tactive ← Tactive − {Tk};
What these code do is removing the track in which happens occlusions between objects form Tactive. And I thought k is the order of tracks, so I can't understand this line: "if k /∈ B", it seems that k is the last coordinate of Tk, is this someting wrong with your denotaion?

(2)line 21 in Algorithm 2: Tracktor algorithm (public detections)
for dt ∈ Dt do
dt, st ← detector.reg and class(dt);
if st < σactive then
Dt ← Dt − {dt};
else
S ← S + {st};
Dt ← NMS(Dt, S, λnew);
I noticed that you regeress and reclassify every element in Dt, is this step necessary? Because I think when we get the public detections Dt, we get coordinate and confidence score. Why not just use those?

Thank you in advance!