DAC-DETR: Divide the Attention Layers and Conquer

This paper reveals a characteristic of DEtection Transformer (DETR) that negatively impacts its training efficacy, i.e., the cross-attention and self-attention layers