AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

Language-driven Fine-grained Retrieval

Shijie Wang, Xin Yu, Yadan Luo, Zijian Wang, Pengfei Zhang, Zi Huang

CVPR 2026 · Vision-language and multimodal › Human and embodied modeling · approach: Composed and text-based image retrieval · 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 EMA-updated projectors for alignment loss instead of backprop-trained projectors (ablation)LaFG with Laux · CUB-200-2011Recall@1: 85.3% -> 87.2%Improved
add language-driven alignment loss to Laux (ablation)LaFG · CUB-200-2011Recall@1: 82.6% -> 87.2%Improved
use LLM-generated attribute descriptions instead of hand-crafted template (ablation)LaFG · CUB-200-2011Recall@1: 83.7% -> 85.3%Improved
add clustered attribute vocabulary and category-specific attribute selection (ablation)LaFG with LLM descriptions · CUB-200-2011Recall@1: 85.3% -> 87.2%Improved
increase LLM descriptions per class from 20 to 25 (ablation)LaFG · CUB-200-2011Recall@1: 87.2% -> 86.3%Hurt
increase LLM descriptions per class from 5 to 20 (ablation)LaFG · CUB-200-2011Recall@1: 81.4% -> 87.2%Improved
increase vocabulary size from 128 to 512 (ablation)LaFG · CUB-200-2011Recall@1: 87.2% -> 85.9%Hurt
reduce vocabulary size from 128 to 32 (ablation)LaFG · CUB-200-2011Recall@1: 87.2% -> 84.3%Hurt
increase Top-K attributes from 40 to 48 (ablation)LaFG · CUB-200-2011Recall@1: 87.2% -> 86.9%Hurt
increase projector distribution dimension from 256 to 512 (ablation)LaFG · CUB-200-2011Recall@1: 87.2% -> 86.9%Hurt
vary EMA update ratio alpha from 0.2 to 0.1 (ablation)LaFG · CUB-200-2011Recall@1: 87.2% -> 87.1%No clear effect
increase EMA update ratio to 0.6 or 0.8 (ablation)LaFG · CUB-200-2011Recall@1: 87.2% -> 86.6%Hurt

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