AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

Beyond [cls]: Exploring the True Potential of Masked Image Modeling Representations

Marcin Przewięźlikowski, Randall Balestriero, Wojciech Jasiński, Marek Śmieja, Bartosz Zieliński

ICCV 2025 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Masked autoencoder pretraining · 8 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 [cls] token with Selective Aggregation (AbMILP) of patch tokensMAE ViT-B · ImageNet-1k linear probetop-1 accuracy: 67.8 -> 71.6Improved
replace [cls] with Selective Aggregation on SimMIMSimMIM ViT-B · ImageNet-1k linear probetop-1 accuracy: 51.5 -> 62.8Improved
replace average pooling with Selective Aggregation on CAPICAPI ViT-L · ImageNet-1k linear probetop-1 accuracy: 76.2 -> 82.4Improved
apply Selective Aggregation to JEA models (DINO, MoCo-v3)DINO ViT-B · ImageNet-1k linear probetop-1 accuracy: 76.6 -> 75.2No clear effect
apply Selective Aggregation to fine-tuned MAEMAE ViT-B fine-tuned · ImageNet-1k linear probetop-1 accuracy: 80.0 -> 79.1Hurt
also aggregate [cls] token with patches (ablation)Selective Aggregation on MIMs · ImageNet-1k linear probetop-1 accuracy: 71.6 -> 71.5 (MAE ViT-B)No clear effect
apply Selective Aggregation for fine-grained classificationMAE ViT-B · CUB-200accuracy: 45.8 -> 65.9Improved
apply Selective Aggregation on BEIT-v2BEIT-v2 ViT-B · OxfordIIIPetsaccuracy: 93.7 -> 93.4No clear effect
replace Tanh with ReLU in AbMILP MLP scorer (ablation)AbMILP aggregation · ImageNet-1k linear probetop-1 accuracy: 68.7 -> 71.7Improved
use deeper MLP scorer instead of linear scorer in AbMILP (ablation)AbMILP aggregation · ImageNet-1k linear probetop-1 accuracy: 71.6 -> 71.5No clear effect
use learned AbMILP instead of non-trainable attention-based aggregation (ablation)MAE ViT-B · ImageNet-1k linear probetop-1 accuracy: 65.2 -> 71.6Improved
use 12-head Attentive Probing instead of AbMILP (ablation)MAE ViT-B · ImageNet-1k linear probetop-1 accuracy: 71.6 -> 75.4Improved

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