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

[Help requested] Clarification on Generating Customized Positive Queries. #147

Open
gracikk-ds opened this issue Jun 24, 2024 · 1 comment

Comments

@gracikk-ds
Copy link

gracikk-ds commented Jun 24, 2024

Hi, guys! Great work!

@TempleX98, I have been studying your paper on "DETRs with Collaborative Hybrid Assignments Training" and I am particularly interested in the Customized Positive Queries Generation process.
In DAB-DETR, there are two types of queries used by the decoder: content queries and positional queries. Please confirm if my understanding is correct on how to generate these queries when integrating your method into DAB-DETR. And even more, do I need to use DAB DETR with content and positional queries?

Here is my understanding:

  • Extract positive coordinates from the auxiliary heads to identify likely object locations.
  • Take those coordinates as positional queries of shape [x, y, w, h].
  • Generate Content Queries by extracting corresponding features from the encoder’s output at these positive coordinates and transforming them appropriately.
  • Combine these content and positional queries to form the input queries for the decoder.

I would appreciate your confirmation or any additional insights on this process.
Thank you for your assistance.

Best regards,
Aleksandr Gordeev

@TempleX98
Copy link
Collaborator

Similar to Co-Deformable-DETR, the content queries are obtained via an MLP layer.

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