AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Saliency-Driven Token Merging for Vision Transformers

Weiying Xie, Xiaoyu Chen, Xin Zhang, Chenhe Hao, Jitao Ma, Yunsong Li, Leyuan Fang

CVPR 2026 · Architectures and efficiency › Attention and transformers · approach: Efficient and robust vision transformers

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
combine saliency outlier score with class attention (alpha=beta=0.5) instead of saliency alone (ablation)SAD-TM on DeiT-Small · ImageNet-1k, 2.6 GFLOPsTop-1 accuracy: 78.02 -> 79.27Improved
combine class attention with saliency instead of attention alone (ablation)SAD-TM on DeiT-Base · ImageNet-1k, 10.1 GFLOPsTop-1 accuracy: 80.48 -> 80.60Improved
combine class attention with saliency instead of attention alone (ablation)SAD-TM on DeiT-Tiny · ImageNet-1k, 0.8 GFLOPsTop-1 accuracy: 71.47 -> 71.58Improved
add deferred merging (SAD-TM-DM)SAD-TM on DeiT-Small · ImageNet-1k, 3.0 GFLOPsTop-1 accuracy: 79.52 -> 79.71Improved
add deferred merging (SAD-TM-DM)SAD-TM on MAE-Huge · ImageNet-1k, ~41% FLOPs reductionTop-1 accuracy: 86.43 -> 86.58Improved
add deferred merging (SAD-TM-DM)SAD-TM on LV-ViT-S · ImageNet-1k, 3.9 GFLOPsTop-1 accuracy: 82.45 -> 82.60Improved
use saliency outlier scores computed by a different DeiT model (ablation)SAD-TM on DeiT-Tiny/Small/Base · ImageNet-1kTop-1 accuracy: at most 0.12% differenceNo clear effect
replace Pearson correlation with cosine similarity (ablation)SAD-TM on DeiT-Base · ImageNet-1k, 10.1 GFLOPsTop-1 accuracy: 80.60 -> 80.44Hurt
replace Pearson correlation with Euclidean distance or Mahalanobis distance (ablation)SAD-TM on DeiT-Small · ImageNet-1k, 2.5 GFLOPsTop-1 accuracy: 79.15 -> 79.10 / 79.11No clear effect
replace Pearson correlation with Mahalanobis distance (ablation)SAD-TM on DeiT-Tiny · ImageNet-1k, 0.8 GFLOPsTop-1 accuracy: 71.58 -> 71.61No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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