REViT-v2: Hierarchical Windowed Roto-reflection Equivariant ViT for Equivariant Feature Extraction

Sheir A. Zaheer, Jihwan Moon, Chan Y. Park
10/6/2026
cs.CVcs.LG

Official code: kc-ml2/revit (6 stars, Apache-2.0, last commit 2026-09-07).

Abstract

We propose a scalable roto-reflection-group-equivariant vision transformer based on windowed group-convolutional self-attention and a hierarchical feature architecture. We demonstrate that our approach can be scaled to group-equivariant vision transformers (ViTs) with millions of parameters and large datasets with practically sized images, i.e., ImageNet. The code and pretrained weights for the proposed Hierarchical Windowed Roto-reflection Equivariant ViTs (REViT-v2) are available at https://github.com/kc-ml2/revit.

View on arXivView PDF

Code Implementations(1)

kc-ml2/revitOfficial100%
Apache-2.0

REViT is the official repository for ICML 2026 paper titled "REViT: Roto-reflection Equivariant Convolutional Vision Transformer"

60PythonMay 22, 20261 months agoApache-2.0

Similar Papers

REViT: Roto-reflection Equivariant Convolutional Vision Transformer

88% match1 repo62026

Vanilla Group Equivariant Vision Transformer: Simple and Effective

83% match2026

Low-Resolution Self-Attention for Semantic Segmentation

80% match1 repo792023

CrossFormer++: A Versatile Vision Transformer Hinging on Cross-scale Attention

79% match1 repo4042023

IPT-V2: Efficient Image Processing Transformer using Hierarchical Attentions

79% match2024

FViT: A Focal Vision Transformer with Gabor Filter

79% match1 repo192024

Cite this paper

@article{zaheer2026revitv,
  title  = {REViT-v2: Hierarchical Windowed Roto-reflection Equivariant ViT for Equivariant Feature Extraction},
  author = {Sheir A. Zaheer and Jihwan Moon and Chan Y. Park},
  year   = {2026},
  eprint = {2610.07585},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2610.07585v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2610.07585.svg)](https://codeofpaper.com/paper/2610.07585)

Discussion