AI Paper Atlas

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

Unsqueeze [CLS] Bottleneck to Learn Rich Representations

Qing Su, Shihao Ji

ECCV 2024 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Data-efficient and robust self-supervision · 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 DINO/iBOT objectives with UDI (stratified local-prediction consolidation plus multimodal prediction)ViT-S/16 SSL · ImageNet-1K linear probingtop-1 accuracy: 77.4 -> 77.6Improved
use the extra multimodal class token for linear evaluationUDI · ImageNet-1K linear probingtop-1 accuracy: 77.6 -> 75.1Hurt
use the extra multimodal class token for transfer learning and fine-tuningUDI · ImageNet-1K fine-tuningtop-1 accuracy: 82.6 -> 83.0Improved
add UDI self-attention pooled patch-level semantic constraint to DINO (ablation)DINO ViT-S/16 100 epochs · COCO detectionAP box: 40.6 -> 42.6Improved
add shared projector for image- and patch-level representations (ablation)UDI · ImageNet-1K lineartop-1 accuracy: 75.6 -> 75.8Improved
add shared projector to EsViT constraint (ablation)EsViT · COCO detectionAP box: 41.8 -> 41.2Hurt
add multimodal objective L_cls+ (ablation)UDI · ImageNet-1K k-NNk-NN top-1: 73.0 -> 74.9Improved
extra class token without multimodal objective (register token only) (ablation)UDI · ImageNet-1K lineartop-1 accuracyNo clear effect
increase blending factor alpha from 0.1 to 0.5 (ablation)UDI · ImageNet-1K lineartop-1 accuracy: 75.9 -> 76.3Improved
add UDI to DINO-style pretraining versus SelfPatchViT-S/16 · COCO Mask R-CNNAP box: 42.1 -> 43.2Improved
use UDI with register token (DINO+reg.)DINO · ImageNet-1K lineartop-1 accuracy: 76.1 -> 76.1No 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