Aug 2026
Vision transformers need zoomer: Efficient ViT with categorical attention prior-guided zoom adapter.
Experiments show ZoomViT, based on the DeiT-S framework, achieves 83.8% top-1 accuracy on ImageNet-1k, surpassing existing efficient state-of-the-art (SOTA) ViTs in accuracy and efficiency.
Ziwen Chen, Xiaoyuan Zhang, Yushi Li et al.
· Neural Networks · 0 citations