impiga / Plain-DETR

[ICCV2023] DETR Doesn’t Need Multi-Scale or Locality Design

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Question about the BoxRPB

Adlith opened this issue · comments

Congratulations on the great work.
How can you get the coordinate of K predicted boxes to generate BoxRPB before cross-attention?

We will release the source code soon.

We will release the source code soon.

Could you give us an estimate of when?

@impiga Yutong is currently in the process of finalizing the code and checkpoints. We invite you to stay tuned for updates.

Hi, @Adlith

We adopt the two-stage setting following Deformable DETR. So, the model would generate bounding box proposals prior to executing the Decoder.