AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Random Wins All: Rethinking Grouping Strategies for Vision Tokens

Qihang Fan, Yuang Ai, Huaibo Huang, Ran He

CVPR 2026 · Architectures and efficiency › Attention and transformers · approach: Efficient and robust vision transformers · 0 citations (OpenAlex)

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 Swin window grouping with random grouping (and CPE)Swin-T · ImageNet-1Ktop-1 accuracy: 81.3 -> 82.7Improved
replace Quadtree grouping with random groupingQuadtree-b2 · ImageNet-1Kthroughput: 467 -> 1926 imgs/sImproved
replace Swin grouping with random groupingSwin-T Mask R-CNN · COCO val2017box AP: 43.7 -> 46.0Improved
random grouping in PTv3PTv3 · ScanNet valmIoU / latency: mIoU 77.6 -> 77.8; latency 88 -> 68 msImproved
apply random token-order pattern in LLMLLaVA-1.5 · POPEaccuracy: 86.1 -> 87.4Improved
random pattern on LLaVA-1.5LLaVA-1.5 · VQAv2accuracy: 78.5 -> 78.5No clear effect
keep positional encoding (ablation)Random-Swin-T · ImageNet-1Ktop-1 accuracy: 79.3 -> 82.7Improved
use a different random tensor per head (ablation)Random-Swin-T · ImageNet-1Ktop-1 accuracy: 80.5 -> 82.7Improved
keep global receptive field in random grouping (ablation)Random-DeiT-S · ImageNet-1Ktop-1 accuracy: 79.3 -> 80.9Improved
keep global receptive field in random grouping (ablation)Random-PVTv2-B1 · ImageNet-1Ktop-1 accuracy: 79.3 -> 79.6No clear effect
fix the random grouping pattern across images (ablation)Random-Swin-T · ImageNet-1Ktop-1 accuracy: 76.4 -> 82.7Improved
add fixed pattern, then multi P, then CPE to fully random grouping (ablation)Swin-T · ImageNet-1Ktop-1 accuracy: 71.2 -> 77.6 -> 80.1 -> 82.7Improved

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