-
Notifications
You must be signed in to change notification settings - Fork 5
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
about Dense Query Recollection and Recurrence #12
Comments
I did not test that, but I think it will work on DETR as well . |
Hello! Could you please provide more information or guidance on how to correctly implement this step? Since I trained it on dab-deformable-detr following the instruction of the paper, feeding back outputs of layer 6 to itself again, but the AP result of evaluation when I shared the parameters of layer 6 for all layers was nearly 0. |
Hi, I get a draft implementation (but correct and runable) of DQRR on adamixer In the implementation, fakesetsize is for acceleration and you can ignore it if stage == self.num_stages, i.e., at the latest stage, we treat it differently from the other stages During testing, you can use the last stage only.
|
I'm not sure, but did you make the order of post-norm/pre-norm correctly when you implement the recurrence? |
Hi, thank you for your amazing work.
I am particularly interested in the section on DQRR which your team has implemented on AdamXier. I am curious to know whether your team has also tested this on DETR, and if so, what the results were. Any details you could share would be greatly appreciated.
The text was updated successfully, but these errors were encountered: