ChannelTok: Efficient Flexible-Length Vision Tokenization 文章

ArXiv CS.CV2026-06-04NEWSen作者: Sukriti Paul, Arpit Bansal, Tom Goldstein
查看原文 →
关系图谱

ChannelTok: Efficient Flexible-Length Vision Tokenization · 相关技术

相关技术

stochastic tail-dropping paradigmCNN-Transformer hybrid backbonechannel-wise flexible-length tokenizer