AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Align Once to Explain: Feature Alignment for Scalable B-cosification of Foundational Vision Transformers

Raphael Maser, Siddhartha Gairola, Sukrut Rao, Bernt Schiele

CVPR 2026 · Architectures and efficiency › Attention and transformers · approach: Token merging and masked pretraining · 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 B-cosification with ALOE feature alignmentSigLIP2 ViT-B/16 B-cos · ImageNet-1k linear probetop-1 accuracy (%): 75.84 -> 83.67Improved
replace B-cosification with ALOEDINOv3 ViT-B/16 B-cos · k-NN averageaccuracy (%): 50.77 -> 82.63Improved
replace B-cosification with ALOESigLIP2 ViT-B/16 B-cos · ImageNet-1k zero-shottop-1 accuracy (%): 61.01 -> 77.20Improved
use B-cos ALOE attributions instead of post-hoc AttnLRPSigLIP2 ViT-B/16 · GridPG localizationGrid-PG score (%): 54.4 -> 84.2Improved
add layer-wise alignment at {2L/3, L} to global-only alignment (ablation)ALOE SigLIP2 · 10-dataset linear probe averageaccuracy (%): 75.51 -> 85.24Improved
supervise all layers instead of three depths (ablation)ALOE SigLIP2 · 10-dataset linear probe averageaccuracy (%): 85.42 -> 84.93No clear effect
reduce alignment data from 100% to 1% of YFCC15M (ablation)ALOE SigLIP2 · ImageNet-1k linear probetop-1 accuracy (%): 83.80 -> 83.33No clear effect
ALOE vs teacher on ImageNet-1k linear probeALOE SigLIP2 ViT-B/16 · ImageNet-1k linear probetop-1 accuracy (%): 84.24 -> 83.67Hurt
cosine alignment loss vs MSE/InfoNCE (ablation)ALOE · linear probeaccuracyMixed

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