又一个轻量级 ViT:Lite Vision Transformer with Enhanced Self-Attention
LiteVisionTransformerwithEnhancedSelf-Attention[pdf]Figure1.MobileCOCOpanopticsegmentation.Themodelneedstorecognize,localize,andsegmentbothobjectsandstuffsatthesametime.Allthemethodshavelessthan5.5Mpa