Data Association

[yanghaibin-cool]

@HanGuangXin
Hello, sorry to bother you again！
When the data is associated, why should the high-scoring detection frame that does not match the trajectory for the first time (motion+reid) and the trajectory that does not match the high-scoring detection frame match again through IoU (motion)? What is the benefit of doing this?
Looking forward to your reply.

[HanGuangXin]

show me the code, please.

…

---Original--- From: "poplar ***@***.***> Date: Fri, Apr 15, 2022 11:45 AM To: ***@***.***>; Cc: "Han ***@***.******@***.***>; Subject: [HanGuangXin/ByteTrack_ReID] Data Association (Issue #15) @HanGuangXin Hello, sorry to bother you again！ When the data is associated, why should the high-scoring detection frame that does not match the trajectory for the first time (motion+reid) and the trajectory that does not match the high-scoring detection frame match again through IoU (motion)? What is the benefit of doing this? Looking forward to your reply. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***>

[yanghaibin-cool]

show me the code, please.
…
---Original--- From: "poplar @.> Date: Fri, Apr 15, 2022 11:45 AM To: @.>; Cc: "Han @.@.>; Subject: [HanGuangXin/ByteTrack_ReID] Data Association (Issue #15) @HanGuangXin Hello, sorry to bother you again！ When the data is associated, why should the high-scoring detection frame that does not match the trajectory for the first time (motion+reid) and the trajectory that does not match the high-scoring detection frame match again through IoU (motion)? What is the benefit of doing this? Looking forward to your reply. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

ByteTrack_ReID/yolox/tracker/byte_tracker_FairMOT.py

Lines 262 to 279 in c3deeec

	''' Step 2: First association, with high score detection boxes'''
	detections = [detections[i] for i in u_detection]
	r_tracked_stracks = [strack_pool[i] for i in u_track if strack_pool[i].state == TrackState.Tracked]
	dists = matching.iou_distance(r_tracked_stracks, detections) # IoU distance. 1 - _ious
	if not self.args.mot20:
	dists = matching.fuse_score(dists, detections) # refine 'dists' with detection scores
	matches, u_track, u_detection = matching.linear_assignment(dists, thresh=self.args.match_thresh) # Hungarain
	for itracked, idet in matches: # do update w.r.t matching results
	track = r_tracked_stracks[itracked]
	det = detections[idet]
	if track.state == TrackState.Tracked:
	track.update(detections[idet],
	self.frame_id) # update of class STrack, update Kalman Filter, track state and other settings
	activated_starcks.append(track)
	else:
	track.re_activate(det, self.frame_id, new_id=False)
	refind_stracks.append(track)

[HanGuangXin]

Step 1 in byte_tracker_FairMOT.py only uses appearance for association, so step 2 is necessary.
If you do not understand, please refer to FairMOT paper.

[HanGuangXin]

If you are not familiar with classic algorithms in MOT like SORT/DeepSORT/JDE/FairMOT, I think it is necessary for you to learn them first!

If there is not other questions, I will close this issue.

[yanghaibin-cool]

If you are not familiar with classic algorithms in MOT like SORT/DeepSORT/JDE/FairMOT, I think it is necessary for you to learn them first!

If there is not other questions, I will close this issue.

OK, thanks for the reply. Next, I will study the code carefully.

[972821054]

@yanghaibin-cool 您好，请问您有没有对没有track_id的情况进行处理？麻烦您了！