Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

question about backward_tracking #66

Open
XiaoyuShi97 opened this issue Jan 26, 2024 · 1 comment
Open

question about backward_tracking #66

XiaoyuShi97 opened this issue Jan 26, 2024 · 1 comment

Comments

@XiaoyuShi97
Copy link

https://github.com/16lemoing/dot/blob/e32c6f7de12342460371f2efc4789bd79c4a39a3/dot/models/shelf/cotracker2_utils/predictor.py#L145-L153

Hi, I am a bit confused about backward_tracking. First, back_tracking means computing the trajectory starting from the last frame to the first N-1 frames? If so, does backward_tracking benefit (get higher accuracy) from forward_tracking? And do the variables "tracks and visibilities" in L153 overwrite those in L145? Thanks!

@nikitakaraevv
Copy link
Contributor

Hi @XiaoyuShi97, CoTracker is an online algorithm, and can only track forward. That's why there's such a thing as backward tracking, where we simply run tracking forward on a reversed video.

The variables "tracks and visibilities" in L153 overwrite those in L145 only for the frames for which the model didn't predict any tracks. For example, if a point was queried at a frame number 10, forward tracking will predict tracks for frames 10,11,12,..., while backward tracking will add predictions for 9,8,7,...

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants