AI Paper Atlas

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

Task Alignment: A simple and effective proxy for model merging in computer vision

Pau de Jorge Aranda, César DE SOUZA, Björn Michele, Mert Bulent SARIYILDIZ, Philippe Weinzaepfel, Florent Perronnin, Diane Larlus, Yannis Kalantidis

ECCV 2026 · Vision-language and multimodal › CLIP and zero-shot · approach: Fine-grained improvements to CLIP

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
select hyperparameters with TAP instead of labeled evaluationTask Arithmetic · CLIP classification (avg over ViT-B-32/ViT-L-14, T=8/14/20)average accuracy: 72.7 -> 72.5No clear effect
select hyperparameters with TAP instead of labeled evaluationIso-CTS · CLIP classification (avg over ViT-B-32/ViT-L-14, T=8/14/20)average accuracy: 86.9 -> 86.8No clear effect
select hyperparameters with TAP instead of labeled evaluationTSV · CLIP classification (avg over ViT-B-32/ViT-L-14, T=8/14/20)average accuracy: 85.2 -> 85.1No clear effect
replace entropy loss with TAP in AdaMergingAdaMerging · CLIP classification (avg over ViT-B-32/ViT-L-14, T=8/14/20)average accuracy: 81.5 -> 81.1Mixed
merge fine-tuned LiDAR encoders with TAP-selected hyperparametersAdaMerging · LiDAR segmentation (4 datasets)normalized performance: 0.841 -> 0.985Improved
merge task-specific encoders on top of DUNE with TAPAdaMerging · DUNE heterogeneous tasks (base DUNE)normalized performance: 0.929 -> 0.971Improved
merge with TIES instead of Task Arithmetic on DUNE tasksTIES · DUNE heterogeneous tasks (base DUNE)normalized performance: 0.956 -> 0.934Hurt
merge DINOv2 fine-tuned models with TAP instead of distilled DUNEIso-CTS · DUNE heterogeneous tasks (base DINOv2)normalized performance: 0.847 -> 0.915Improved
use TAP selection to cut hyperparameter search costTAP · LiDAR segmentationgpu time per candidate: 30 gpu hours -> 7 minutesImproved
vary number of images used to compute TAP (very few images) (ablation)TAP · DUNE tasksselected modelNo clear effect
vary feature distance (l1, l2, cosine) in TAP (ablation)TAP · DUNE tasksTAP curveNo 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