AI Paper Atlas

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

ExpAlign: Expectation-Guided Vision–Language Alignment for Open-Vocabulary Grounding

Junyi Hu, Tian Bai, Fengyi Wu, Wenyan Li, Zhenming Peng, Yi ZHANG

ICML 2026 · Vision-language and multimodal › CLIP and zero-shot · approach: Open-vocabulary alignment from VLMs · 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
use token-level alignment (EAH) instead of mean pooling (ablation)ExpAlign · LVIS minivalAP: 31.9 -> 37.1Improved
use token-level alignment (EAH) instead of EOT pooled token (ablation)ExpAlign · LVIS minivalAP: 34.4 -> 37.1Improved
use token-level alignment (EAH) instead of mean pooling (ablation)ExpAlign · LVIS minivalrare AP: 27.3 -> 36.2Improved
add semantic contrastive loss weight 0.5 (ablation)ExpAlign · LVIS minivalAP: 35.0 -> 37.0Improved
add geometry-aware consistency (GACO) weight 0.5 to semantic loss 0.5 (ablation)ExpAlign · LVIS minivalAP: 37.0 -> 37.8Improved
raise semantic loss weight to 1.0 with geo weight 0.5 (ablation)ExpAlign · LVIS minivalAP: 37.8 -> 36.2Hurt
increase geo weight from 1.0 to 1.5 with sem 0.5 (ablation)ExpAlign · LVIS minivalAP: 37.2 -> 36.7Hurt
use frozen DINOv3 backbone instead of YOLOv8 (ablation)ExpAlign · LVIS minivalAP: 35.6 -> 37.2Improved
freeze the DINOv3 backbone (ablation)ExpAlign · LVIS minivalAP: N/A -> 37.2Improved
add RefCOCO data to pre-trainingExpAlign · RefCOCO valtop-1 accuracy: 42.2 -> 51.6Improved
add RefCOCO data to pre-trainingExpAlign · LVIS valAP: 30.3 -> 29.5Hurt
add RefCOCO data to pre-trainingExpAlign · LVIS minivalAP: 37.2 -> 37.1No 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