AI Paper Atlas

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

Difference Vector Equalization for Robust Fine-tuning of Vision-Language Models

Satoshi Suzuki, Shin'ya Yamaguchi, Shoichiro Takeda, Taiga Yamane, Naoki Makishima, Naotaka Kawata, Mana Ihori, Tomohiro Tanaka, Shota Orihashi, Ryo Masumura

AAAI 2026 · Vision-language and multimodal › CLIP and zero-shot · approach: Feature calibration for unseen-class detection · 1 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
add AVL (average vector loss) to FLYP (ablation)FLYP · ImageNet OOD avgtop-1 accuracy: 59.5 -> 62.9Improved
add AVL (average vector loss) to FLYP (ablation)FLYP · ImageNet zero-shot avgtop-1 accuracy: 49.5 -> 62.9Improved
add PVL (pairwise vector loss) to FLYP (ablation)FLYP · ImageNet zero-shot avgtop-1 accuracy: 49.5 -> 62.7Improved
add PVL on top of AVL (ablation)DiVE · ImageNet OOD avgtop-1 accuracy: 62.9 -> 63.2Improved
add PVL on top of AVL (ablation)DiVE · ImageNet zero-shot avgtop-1 accuracy: 62.9 -> 63.7Improved
add AVL/PVL to FLYPFLYP · ImageNet IDtop-1 accuracy: 82.2 -> 82.5Improved
use EMA average vector (alpha 0.99) instead of per-batch average (alpha 0) (ablation)AVL · ImageNet zero-shot avgtop-1 accuracy: 61.4 -> 62.9Improved
constrain difference vectors instead of cosine similarities (ablation)DiVE · ImageNet zero-shot avgtop-1 accuracy: 61.7 -> 63.7Improved
use larger reference dataset (CC3M vs Flickr8K) (ablation)DiVE · ImageNet zero-shot avgtop-1 accuracy: 55.2 -> 63.7Improved
use DiVE instead of ARF on ImageNetARF · ImageNet zero-shot avgtop-1 accuracy: 55.6 -> 63.7Improved
use DiVE instead of ARF on ImageNet IDARF · ImageNet IDtop-1 accuracy: 82.7 -> 82.5Hurt
allow non-zero difference vectors (DiVE) instead of zero (SnD)SnD · ImageNet OOD avgtop-1 accuracy: 60.7 -> 63.2Improved

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