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

Extract same number of regions. #14

Open
chrisrn opened this issue Nov 28, 2016 · 5 comments
Open

Extract same number of regions. #14

chrisrn opened this issue Nov 28, 2016 · 5 comments

Comments

@chrisrn
Copy link

chrisrn commented Nov 28, 2016

Can I extract exactly 20 bounding boxes for every image? I changed the scale parameter but the number of bounding boxes differs between different images.

@UpCoder
Copy link

UpCoder commented Aug 11, 2017

Did you solved the problem? I also want to determine the number of bounding box.

@sdadia
Copy link

sdadia commented Aug 12, 2017

  • You cannot control the exact number of extracted box directly.

  • But you can just select top 20 boxes using :

top_20_boxes = regions[0:20]
  • But you can just select top N boxes using :
top_N_boxes = regions[0:N]

@UpCoder
Copy link

UpCoder commented Aug 12, 2017

Thanks for replying. I am reading the paper R-CNN, In this paper, the author said they extract 2000 region from every picture. They also using selective region method to extract region. So I feel confused, I thought the selective region method can control the number of bounding box.

@sdadia
Copy link

sdadia commented Aug 12, 2017

The selective search algorithm does not permit exact control of boxes (read the original selective search paper) nor any variants of selective search.

The R-CNN authors first do non-maxima suppression and then simply just pick top 2000 boxes. This leads to boxes which are non-overlapping and boxes are spread around the whole image minimizing the chance to missed an object.

@UpCoder
Copy link

UpCoder commented Aug 12, 2017

em, I will continue to understand this paper, I not yet totally understand that. Thank you very much.

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