AI Paper Atlas

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

FG-CLIP 2: A Bilingual Fine-grained Vision-Language Alignment Model

Chunyu Xie, Bin Wang, Fanjing Kong, Jincheng Li, Dawei Liang, Ji Ao, Dawei Leng, Yuhui Yin

ICML 2026 · Vision-language and multimodal › Human and embodied modeling · approach: Scene graph generation · 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
add Textual Intra-modal Contrastive (TIC) loss (ablation)FG-CLIP 2 ViT-B/16 · COCO80 bounding box classificationtop-1 accuracy: 70.1 -> 74.9Improved
add TIC loss (ablation)FG-CLIP 2 ViT-B/16 · FG-OVD Hardtop-1 accuracy: 51.6 -> 52.3Improved
add cross-modal rank loss with global threshold synchronization (ablation)FG-CLIP 2 ViT-B/16 · FG-OVD Hardtop-1 accuracy: 50.9 -> 52.3Improved
add cross-modal rank loss (ablation)FG-CLIP 2 ViT-B/16 · COCO80 bounding box classificationtop-1 accuracy: 74.0 -> 74.9Improved
add both LCMR and LTIC (ablation)FG-CLIP 2 ViT-B/16 · COCO80 bounding box classificationtop-1 accuracy: 62.7 -> 74.9Improved
add LCMR and LTIC (ablation)FG-CLIP 2 ViT-B/16 · ImageNet-V2top-1 accuracy: 72.0 -> 72.2No clear effect
add Chinese data in stage 2 (bilingual vs English-only) (ablation)FG-CLIP 2 · Chinese and English benchmarksretrieval/classificationImproved
replace SigLIP 2 with FG-CLIP 2 encoder in LLaVA-1.5LLaVA-1.5 · GQAaccuracy: 62.0 -> 64.0Improved
replace SigLIP 2 with FG-CLIP 2 in LLaVA-1.5LLaVA-1.5 · MMBench-CN Devaccuracy: 66.2 -> 67.6No clear effect
add fine-grained region training on top of SigLIP 2 weightsFG-CLIP 2 ViT-B/16 · FG-OVD Hardtop-1 accuracy: 24.9 -> 52.3Improved
use FG-CLIP 2 as the alignment model in LLMDet open-vocabulary detectionLLMDet Swin-T · LVIS minivalAP: 47.9 -> 51.6Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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