AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Semantic Equitable Clustering: A Simple and Effective Strategy for Clustering Vision Tokens

Qihang Fan, Huaibo Huang, Mingrui Chen, Ran He

ICCV 2025 · Architectures and efficiency › Attention and transformers · approach: Token merging and masked pretraining · 7 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 window attention with SEC in Swin-TSwin-T · ImageNet-1Ktop-1 accuracy (%): 81.3 -> 83.8Improved
replace window attention with SEC in Swin-SSwin-S · ImageNet-1Ktop-1 accuracy (%): 83.0 -> 85.0Improved
add SEC to DeiT-SDeiT-S · COCO Mask R-CNNbox AP: 44.5 -> 47.7Improved
add SEC to DeiT-SDeiT-S · ADE20KmIoU: 43.0 -> 47.5Improved
add SEC to DeiT-B (fewer FLOPs)DeiT-B · ImageNet-1Ktop-1 accuracy (%): 81.8 -> 82.4Improved
use SEC instead of EViT token pruning in DeiT-SDeiT-S · ImageNet-1Ktop-1 accuracy (%): 79.8 -> 80.5Improved
add SEC as pretrained backbone under SimMIMSwin-B · ImageNet-1Ktop-1 accuracy (%): 84.0 -> 85.9Improved
use SEC connector at 288+1 tokens instead of full 576+1 tokensLLaVA-1.5 · TextVQAaccuracy: 58.2 -> 60.1Improved
use SEC instead of EViT connector at 144+1 tokensLLaVA-1.5 · TextVQAaccuracy: 51.2 -> 56.8Improved
reduce SEC output tokens from 288+1 to 144+1 (ablation)LLaVA-1.5 · TextVQAaccuracy: 60.1 -> 56.8Hurt
use interleaved instead of sequential token merging in SEC connector (144+1) (ablation)LLaVA-1.5 · TextVQAaccuracy: 47.2 -> 56.8Improved
increase tokens per cluster from 98 to 196 (ablation)SECViT-T · ImageNet-1Ktop-1 accuracy (%): 82.7 -> 83.0No clear effect

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