Open caposerenity opened 2 years ago
Hi, thanks for your issue, which is vital to me! It is my mistake and will be fixed as soon as possible. Meanwhile, I suggest you try adding it according to the official code(https://github.com/dddzg/up-detr/blob/be066dfca5e8f1f35a4331490e8239b579bbf497/models/updetr.py#L50).
Thank you @lifuguan for your reply, I've implementated the attention mask. BTW, could you please provide the train_log and experiment results with this codebase for reference?
Hi, @lifuguan, thank you for your mmdetection implentation for UP-DETR, I'm trying to follow your work, but found that some details in the original paper is not implementated in this repo, e.g. the attention mask. So I'm wondering that is there a performance gap between your implementation and that reported in the original paper?