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

demo parameters #25

Closed
pqviet opened this issue Jun 22, 2021 · 5 comments
Closed

demo parameters #25

pqviet opened this issue Jun 22, 2021 · 5 comments

Comments

@pqviet
Copy link

pqviet commented Jun 22, 2021

I have tried the demo (with default parameters) on a very simple video which is tracking only 1 person, and found 2 following problems

  1. If it fails to detect the person at the current frame, it will assign a new track ID at the next frame
  2. Sometimes there are 2 overlapped detection boxes (with different track IDs) with very high IOU at the same place
    How can I fine-tune the default parameters to overcome these problems?
@JialianW
Copy link
Owner

  1. We recommend you use the Crowdhuman pretrained model to test your own video.

  2. Yes, in some cases, it will assign a new ID if it has low matching score with previous targets.

  3. You can use box NMS to filter out highly overlapped boxes.

@pqviet
Copy link
Author

pqviet commented Jun 22, 2021

Where can I change the track matching score and box NMX? Thanks.

@JialianW
Copy link
Owner

You can change the matching score threshold here:

if max_cos >= 0.3 and self.tracklet_ages[max_id - 1] < self.opt.window_size:

For using box NMS, try to add like this --box_nms 0.7 for your commend.

@Moaz-Morsy
Copy link

how can we test segmentaion model? @JialianW

@JialianW
Copy link
Owner

how can we test segmentaion model? @JialianW

See #24 (comment)

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

3 participants