RIFormer：在不使用 Token Mixer 功能的同时保持视觉骨干网络的有效性（英文版）

英文标题：RIFormer: Keep Your Vision Backbone Effective While Removing Token Mixer中文摘要：本文研究如何在去除基本构建模块中的令牌混合器的同时保持视觉骨干的有效性，并提出了一个可行的优化策略，使得我们能够构建一种极其简单的视觉骨干，具有鼓舞人心的性能，同时在推理过程中享受高效性。英文摘要：This paper studies how to keep a vision backbone effective while removingtoken mixers in its basic building blocks. Token mixers, as self-attention forvision transformers (ViTs), are intended to perform information communicationbetween different spatial tokens but suffer from considerable computationalcost and latency