FLASHSWIN: Unlocking Large Windows and Dense Tokens in Swin Vision Transformers with Memory Efficient Attention

作者：Tushar Kataria, Gerald Sabin, Ponnuswamy Sadayappan, Shireen Y. Elhabian · 单位：Scientific Computing and Imaging Institute, University of Utah, Salt Lake City, UT, USA, Kahlert School of Computing, University of Utah, Salt Lake City, UT, USA · 会议/期刊：arXiv preprint · 方向：cs.CV · 发布日期：2026-10-06