Research map / Vision-language and multimodal / Cross-modal learning
Evo-Retriever: LLM-Guided Curriculum Evolution with Viewpoint-Pathway Collaboration for Multimodal Document Retrieval
Weiqing Li, Jinyue Guo, Yaqi Wang, Haiyang Xiao, Yuewei Zhang, Guohua Liu, Hao Henry Wang
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 multi-view alignment (MVA) (ablation) | Net0 baseline · ViDoRe V2 (3B) | nDCG@5: 61.17 -> 62.25 | Improved |
| use downsample-only augmented view instead of dual view (ablation) | MVA · ViDoRe V2 (3B) | nDCG@5: 61.17 -> 59.75 | Hurt |
| use stitched-only view without original view (ablation) | MVA · ViDoRe V2 (3B) | nDCG@5: 61.17 -> 58.59 | Hurt |
| add bidirectional contrastive learning (BCL) with fixed hard negatives (ablation) | Net0 baseline · ViDoRe V2 (3B) | nDCG@5: 61.17 -> 61.84 | Improved |
| combine MVA and BCL (ablation) | Net0 baseline · ViDoRe V2 (3B) | nDCG@5: 61.17 -> 62.39 | Improved |
| use fixed window 80-98% negatives (ablation) | Net0 baseline · ViDoRe V2 (3B) | nDCG@5: 61.17 -> 62.10 | Improved |
| use dynamic rule-based oracle curriculum instead of fixed window (ablation) | Fixed Window 80-98% · ViDoRe V2 (3B) | nDCG@5: 62.10 -> 62.81 | Improved |
| use LLM-EC instead of rule-based oracle (ablation) | Rule-based Oracle · ViDoRe V2 (3B) | nDCG@5: 62.81 -> 63.05 | Improved |
| add exploration phase (ablation) | LLM-EC · ViDoRe V2 (3B) | nDCG@5: 61.86 -> 63.05 | Improved |
| use 16 intervals instead of 21 intervals (ablation) | LLM-EC · ViDoRe V2 (3B) | nDCG@5: 62.15 -> 63.05 | Improved |
| use Qwen3-32B controller instead of Qwen3-235B-A22B (ablation) | LLM-EC · ViDoRe V2 (3B) | nDCG@5: 63.05 -> 63.30 | No clear effect |
| use margin-loss temperature 0.02 instead of 1.0 (ablation) | Evo-Retriever 3B · ViDoRe V2 (3B) | nDCG@5: 57.37 -> 63.05 | Improved |
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.
- colpali-v1.3 — ColPali: Efficient Document Retrieval with Vision Language Models (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
- Recurrence-Enhanced Vision-and-Language Transformers for Robust Multimodal Document Retrieval (CVPR 2025 · similarity 0.88)
- ITO: Multi-View Alignment and Training-Time Fusion for Image-Text Pretraining (NeurIPS 2026 · similarity 0.87)
- AlignVLM: Bridging Vision and Language Latent Spaces for Multimodal Document Understanding (NeurIPS 2025 · similarity 0.87)
- LamRA: Large Multimodal Model as Your Advanced Retrieval Assistant (CVPR 2025 · similarity 0.87)
- Beyond Global Similarity: Multi-Conditional Retrieval for Fine-Grained Cross-Modal Understanding (CVPR 2026 · similarity 0.86)
- Incorporating Dense Knowledge Alignment into Unified Multimodal Representation Models (CVPR 2025 · similarity 0.86)
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.
