Research map / Vision-language and multimodal / CLIP and zero-shot
ReMatch: Boosting Representation through Matching for Multimodal Retrieval
Qianying Liu, Xiao Liang, Zhiqiang Zhang, Yibo Chen, Xu Tang, Zhongfei Qing, Fengfan Zhou, Yao Hu, Paul Henderson
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 |
|---|---|---|---|
| tune training setup (LoRA, LR, cosine schedule, temperature) (ablation) | VLM2Vec baseline · MMEB overall | Hit@1: 59.7 -> 65.5 | Improved |
| add target instructions (ablation) | tuned baseline · MMEB overall | Hit@1: 65.5 -> 67.3 | Improved |
| add one explicit hard negative (ablation) | tuned baseline · MMEB overall | Hit@1: 67.3 -> 67.9 | Improved |
| add multi-view query-document matching (MVQDM) (ablation) | Exp3 baseline · MMEB overall | Hit@1: 67.9 -> 68.4 | Improved |
| add chat-style template to embedding extraction (MVQDM++) (ablation) | Exp3 baseline · MMEB overall | Hit@1: 67.9 -> 68.8 | Improved |
| add multi embedding fusion with learnable tokens (ablation) | Exp3 baseline · MMEB overall | Hit@1: 67.9 -> 68.7 | Improved |
| combine MVQDM++ and MEF (ablation) | Exp3 baseline · MMEB overall | Hit@1: 67.9 -> 69.2 | Improved |
| use all raw/feature views instead of Feat <-> Feat only (ablation) | MVQDM · MMEB overall | Hit@1: 68.1 -> 68.4 | Improved |
| raise matching loss weight from 0.1 to 0.4 (ablation) | MVQDM · MMEB overall | Hit@1: 68.4 -> 68.0 | Hurt |
| average 32 learnable tokens without orthogonal regularization (ablation) | Exp3 baseline · MMEB overall | Hit@1: 67.9 -> 68.2 | No clear effect |
| add orthogonal regularization at 64 tokens (ablation) | Avg fusion · MMEB overall | Hit@1: 68.5 -> 68.8 | Improved |
| add orthogonal regularization at 4 tokens (ablation) | Avg fusion · MMEB overall | Hit@1: 68.6 -> 68.3 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- UniME-V2 — UniME-V2: MLLM-as-a-Judge for Universal Multimodal Embedding Learning (AAAI 2026) · better on 1 of 1 shared result
- MM-EMBED — MM-EMBED: UNIVERSAL MULTIMODAL RETRIEVAL WITH MULTIMODAL LLMS (ICLR 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
- Your Embedding Model Is SMARTer Than You Think (NeurIPS 2026 · similarity 0.89)
- Beyond Global Similarity: Multi-Conditional Retrieval for Fine-Grained Cross-Modal Understanding (CVPR 2026 · similarity 0.89)
- MetaEmbed: Scaling Multimodal Retrieval at Test-Time with Flexible Late Interaction (ICLR 2026 · similarity 0.89)
- 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)
- U-MARVEL: Unveiling Key Factors for Universal Multimodal Retrieval via Embedding Learning with MLLMs (ICLR 2026 · similarity 0.89)
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.
