AI Paper Atlas

Research map / Vision-language and multimodal / CLIP and zero-shot

Supervised Classification Heads as Semantic Prototypes: Unlocking Vision-Language Alignment via Weight Recycling

David Méndez, Roberto Confalonieri, Natalia Díaz-Rodríguez

ICML 2026 · Vision-language and multimodal › CLIP and zero-shot · approach: Open-vocabulary alignment from VLMs

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
use cosine similarity to classifier weight rows instead of the linear layer with bias (ablation)ImageNet-21K pretrained encoders · ImageNet-1K valaccuracy: 82.71 -> 82.10 (BEiT)No clear effect
align with classifier head weights instead of averaged image representations of equal count (ablation)MLP alignment BEiT-B/16 · Flickr30K zero-shot retrievali2t mAP: 0.1860 -> 0.2751Improved
align with classifier weights instead of one random image representation per class (ablation)MLP alignment BEiT-B/16 · Flickr30K zero-shot retrievali2t mAP: 0.1311 -> 0.2751Improved
align with classifier weights instead of random image representations (ablation)MLP alignment BEiT-B/16 · Zero-shot classification, 9 datasetsaccuracy: CIFAR100 52.78 -> 61.12Improved
use ImageNet-21K head weights instead of ImageNet-1K weights (ablation)MLP alignment BEiT-B/16 · Zero-shot classification, 9 datasetsaccuracyImproved
add one image-caption pair per class to ImageNet-21K weights (ablation)MLP alignment BEiT-B/16 · Zero-shot classification, 9 datasetsaccuracyMixed
use iNaturalist-2021 weights instead of ImageNet-21K weights (ablation)MLP alignment BEiT-B/16 · Zero-shot classification benchmarksaccuracyHurt
augment image-text pairs with ImageNet-21K head weights (1K pairs) (ablation)CSA · Flickr30K retrievali2t mAP: 0.0668 -> 0.4603Improved
augment image-text pairs with ImageNet-21K head weights (1K pairs) (ablation)MLP alignment · Flickr30K retrievali2t mAP: 0.3702 -> 0.4945Improved
augment image-text pairs with ImageNet-21K head weights (30K pairs) (ablation)Text2cpts · Flickr30K retrievali2t mAP: 0.5895 -> 0.5893No clear effect
augment image-text pairs with ImageNet-21K head weights (30K pairs) (ablation)CSA · Flickr30K retrievalt2i P@1: 0.2432 -> 0.2244Hurt
mitigate modality gap by centering/rescaling or lightweight projections (ablation)weights plus image-text augmentation · downstream tasksperformanceNo 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