Question regarding to trk.hit_streak >= self.min_hits or self.frame_count <= self.min_hits

[zmy1116]

Hello,

First, thank you for sharing this repository.

I'm testing OC-SORT on one of my own video using the demo_video.py file as you suggested in readme. I found that some detection results are not included in any of tracking results. Digging further I found that the main problem is at ocsort.py where we output results:

OC_SORT/trackers/ocsort_tracker/ocsort.py

Lines 316 to 318 in 7a390a5

	if (trk.time_since_update < 1) and (trk.hit_streak >= self.min_hits or self.frame_count <= self.min_hits):
	# +1 as MOT benchmark requires positive
	ret.append(np.concatenate((d, [trk.id+1])).reshape(1, -1))

specifically on line 316 trk.hit_streak >= self.min_hits or self.frame_count <= self.min_hits:

Let's suppose the video has 4 frames, the a still object appear on frames 2, 3, 4, and we set min_hits = 3 as the default params.

• track successfully stablished for the object on frame 2 and 3
• On frame 4: we associate the object, so the hit_streak= 2, min_hits=3, frame_count=4, it does not satisfy the condition to include to the return, hence empty blob is outputted.

I'm a bit confused what's the rational behind trk.hit_streak >= self.min_hits or self.frame_count <= self.min_hits. My guest is that it may boost performance for certain types of videos? Should I use a different function other than update for general videos (eg: not security footage, with shot changes, etc)?

I guess my concern is that I can fix this specific part with something making sense. but there may be other parts you did intentionally for performance on certain types of videos that I'm not aware of.

Thank you

[zmy1116]

i understand now that since it's an online tracker, we only track objects starting from the min_hits