Research map / Vision-language and multimodal / Cross-modal learning
UniME-V2: MLLM-as-a-Judge for Universal Multimodal Embedding Learning
Tiancheng Gu, Kaicheng Yang, Kaichen Zhang, Xiang An, Ziyong Feng, Yueyi Zhang, Weidong Cai, Jiankang Deng, Lidong Bing
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add MLLM-as-a-Judge hard negative mining (ablation) | UniME-V2 (Qwen2-VL-2B) · MMEB | Precision@1: 60.1 -> 61.6 | Improved |
| add soft semantic matching score alignment (ablation) | UniME-V2 (Qwen2-VL-2B) · MMEB | Precision@1: 61.6 -> 63.6 | Improved |
| add soft semantic matching score alignment (ablation) | UniME-V2 (Qwen2-VL-2B) · SugarCrepe compositional | Recall@1: 63.7 -> 64.1 | No clear effect |
| use Qwen2.5-VL-7B instead of InternVL3-8B as judge (ablation) | UniME-V2 (Qwen2-VL-2B) · MMEB | Precision@1: 58.5 -> 63.6 | Improved |
| increase hard negatives from 4 to 8 (ablation) | UniME-V2 (Qwen2-VL-2B) · MMEB | Precision@1: 61.3 -> 63.6 | Improved |
| increase hard negatives from 8 to 10 (ablation) | UniME-V2 (Qwen2-VL-2B) · MMEB | Precision@1: 63.6 -> 63.0 | Hurt |
| UniME-V2 instead of UniME on text-to-image retrieval | UniME-V2 · Flickr30K text-to-image | Recall@1: 94.4 -> 93.7 (LLaVA-OV-7B) | No clear effect |
| UniME-V2 instead of UniME on compositional retrieval | UniME-V2 (LLaVA-OV-7B) · SugarCrepe Replace | Recall@1: 80.5 -> 88.6 | Improved |
| add UniME-V2-Reranker instead of LamRA | UniME-V2 7B retrieval · SugarCrepe compositional | Recall@1: 87.4 -> 94.8 | Improved |
| add UniME-V2-Reranker | UniME-V2 (Qwen2-VL-7B) · MMEB | Precision@1: 68.0 -> 69.6 | Improved |
| UniME-V2 vs UniME on overall MMEB with LLaVA-OV-7B | UniME-V2 · MMEB | Precision@1: 70.7 -> 71.2 | Improved |
Later papers that compared with this method
3 later accepted papers report UniME-V2 as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Magic-MM-Embedding: Towards Visual-Token-Efficient Universal Multimodal Embedding with MLLMs (ECCV 2026 · better on 3 of 3 shared results)
- Illuminating Visual Identity in Universal Multimodal Embeddings (CVPR 2026 · better on 2 of 2 shared results)
- ReMatch: Boosting Representation through Matching for Multimodal Retrieval (CVPR 2026 · better on 1 of 1 shared result)
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
- U-MARVEL: Unveiling Key Factors for Universal Multimodal Retrieval via Embedding Learning with MLLMs (ICLR 2026 · similarity 0.91)
- MM-EMBED: UNIVERSAL MULTIMODAL RETRIEVAL WITH MULTIMODAL LLMS (ICLR 2025 · similarity 0.91)
- VLM2Vec: Training Vision-Language Models for Massive Multimodal Embedding Tasks (ICLR 2025 · similarity 0.90)
- Bridging Modalities: Improving Universal Multimodal Retrieval by Multimodal Large Language Models (CVPR 2025 · similarity 0.88)
- Incorporating Dense Knowledge Alignment into Unified Multimodal Representation Models (CVPR 2025 · similarity 0.88)
- FreeRet: MLLMs as Training-Free Retrievers (ICML 2026 · similarity 0.88)
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.
