Hi, I have been studing Deformable DETR since it was proposed.
The result of Deformable DETR on the SDOA-D is the lowest. In my experiance, the mAP of Deformable DETR should be higher than faster rcnn, mask rcnn.
I have read your config. It seems that the multi-scale training pipeline for DETR has been deleted. It may serously affect the performance of DETR-like model.
Hi, I have been studing Deformable DETR since it was proposed. The result of Deformable DETR on the SDOA-D is the lowest. In my experiance, the mAP of Deformable DETR should be higher than faster rcnn, mask rcnn. I have read your config. It seems that the multi-scale training pipeline for DETR has been deleted. It may serously affect the performance of DETR-like model.