AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Quick ViTs: Speeding up Vision Transformers through Equivariance

David Nordström, Johan Edstedt, Fredrik Kahl, Georg Bökman

ECCV 2026 · Architectures and efficiency › Attention and transformers · approach: Hybrid and interpretable ViT designs

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace blocks with octic H8 layersViT-H/14 DeiT III · ImageNet-1KTop-1: 84.6 -> 85.0Improved
replace blocks with octic I8 layersViT-H/14 DeiT III · ImageNet-1KTop-1: 84.6 -> 84.7No clear effect
use I8 invariant modelViT-H/14 DeiT III · ImageNet-1K rotated valOOD rotation delta accuracy: -12.6 -> 0.0Improved
octic layers instead of standard for DINOv2ViT-H/16 · ImageNet-1K linearaccuracy: 81.7 -> 82.2Improved
I8 octic layers instead of standardViT-H/16 DINOv2 · ADE20K linearmIoU: 34.7 -> 33.9Hurt
I8 octic layers instead of standardViT-L/16 DINOv2 · VOC2012 k-NNmIoU: 58.1 -> 59.6Improved
use equivariant instead of arbitrary patch embedding (ablation)H8(ViT-B) · ImageNet-1K DeiT IIITop-1: 82.4 -> 83.0Improved
use power spectrum invariantization instead of triple correlation (ablation)D8(ViT-L/16) · ImageNet-1KTop-1: 76.8 -> 79.5Improved
use power spectrum instead of max filtering (ablation)D8(ViT-L/16) · ImageNet-1KTop-1: 77.9 -> 79.5Improved
use power spectrum instead of linear invariant (ablation)D8(ViT-L/16) · ImageNet-1KTop-1: 79.2 -> 79.5Improved
keep late blocks non-equivariant (H8 vs I8) (ablation)ViT-H/16 DINOv2 · ImageNet-1K linearaccuracy: 81.9 -> 82.2Improved
scale width from ViT-L to ViT-22B (ablation)octic ViTs · speed bf16speedup: 1.32x -> 3.54xImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy