D^2ETR: Decoder-Only DETR with Computationally Efficient Cross-Scale Attention

Lin, Junyu, Mao, Xiaofeng, Chen, Yuefeng · arXiv · 2022