DEYO: DETR with YOLO for End-to-End Object Detection

Haodong Ouyang

The training paradigm of DETRs is heavily contingent upon pre-training their backbone on the ImageNet dataset. However, the limited supervisory signals provided by the image classi…