AI RESEARCH
ViT-AdaLA: Adapting Vision Transformers with Linear Attention
arXiv CS.CV
•
ArXi:2603.16063v1 Announce Type: new Vision Transformers (ViTs) based vision foundation models (VFMs) have achieved remarkable performance across diverse vision tasks, but suffer from quadratic complexity that limits scalability to long sequences. Existing linear attention approaches for ViTs are typically trained from scratch, requiring substantial computational resources, while linearization-based methods developed for large language model decoders do not transfer well to ViTs.