Research map / Vision-language and multimodal / Cross-modal learning
ELVA: Exploring Ranking-Driven Universal Multimodal Retrieval
Yuhan Liu, Pei Fu, Hang Li, Yukun Qi, Chao Jiang, Jingwen Fu, Zhen Liu, Bin Qin, Zhenbo Luo, Jian Luan, Jingmin Xin
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 ranking reward (ablation) | ELVA-7B · M-BEIR image-to-text subset average | Recall: 57.2 -> 58.7 | Improved |
| add margin reward (ablation) | ELVA-7B · M-BEIR image-to-text subset average | Recall: 58.1 -> 58.7 | Improved |
| add negative-ranking term to the ranking reward (ablation) | ELVA-7B · M-BEIR image-to-text subset average | Recall: 58.1 -> 58.7 | Improved |
| add balanced negative sampling strategy (ablation) | ELVA-7B · M-BEIR image-to-text subset average | Recall: 58.2 -> 58.7 | Improved |
| add randomly sampled negatives (ablation) | ELVA-7B · M-BEIR image-to-text subset average | Recall: 58.4 -> 58.7 | Improved |
| weight ranking reward higher (alpha 0.4, epsilon 0.6 vs 0.6/0.4) (ablation) | ELVA-7B · M-BEIR image-to-text subset average | Recall: 58.2 -> 58.7 | Improved |
| weight ranking reward higher (alpha 0.4, epsilon 0.6 vs 0.5/0.5) (ablation) | ELVA-7B · M-BEIR image-to-text subset average | Recall: 58.4 -> 58.7 | Improved |
| apply ELVA ranking RL to PUMA | PUMA · M-BEIR | average Recall: 54.4 -> 56.3 | Improved |
| apply ELVA ranking RL to MM-Embed | MM-Embed · M-BEIR | average Recall: 52.7 -> 54.9 | Improved |
| apply ELVA ranking RL to PUMA | PUMA · MRBench | average Recall: 35.1 -> 37.0 | Improved |
| apply ELVA ranking RL to MM-Embed | MM-Embed · MRBench | average Recall: 34.6 -> 36.2 | Improved |
| ELVA RL stage widens embedding distance when a grain is masked | ELVA · MRBench | cosine distance: 0.07 -> 0.15 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- MM-Embed-7B — MM-EMBED: UNIVERSAL MULTIMODAL RETRIEVAL WITH MULTIMODAL LLMS (ICLR 2025) · better on 3 of 3 shared results
- LamRA-7B — LamRA: Large Multimodal Model as Your Advanced Retrieval Assistant (CVPR 2025) · better on 1 of 1 shared result
- VLM-R1-7B — VLM-R³: Region Recognition, Reasoning, and Refinement for Enhanced Multimodal Chain-of-Thought (NeurIPS 2025) · 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
- MM-EMBED: UNIVERSAL MULTIMODAL RETRIEVAL WITH MULTIMODAL LLMS (ICLR 2025 · similarity 0.89)
- LamRA: Large Multimodal Model as Your Advanced Retrieval Assistant (CVPR 2025 · similarity 0.89)
- Beyond Global Similarity: Multi-Conditional Retrieval for Fine-Grained Cross-Modal Understanding (CVPR 2026 · similarity 0.89)
- U-MARVEL: Unveiling Key Factors for Universal Multimodal Retrieval via Embedding Learning with MLLMs (ICLR 2026 · similarity 0.89)
- FreeRet: MLLMs as Training-Free Retrievers (ICML 2026 · similarity 0.88)
- Bridging Modalities: Improving Universal Multimodal Retrieval by Multimodal Large Language Models (CVPR 2025 · 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.
