Research map / Vision-language and multimodal / Human and embodied modeling
Contextual Similarity Aggregation with Self-attention for Visual Re-ranking
Jianbo Ouyang, Hui Wu, Min Wang, Wengang Zhou, Houqiang Li
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 |
|---|---|---|---|
| re-rank top-K=1024 candidates instead of no re-ranking | R-GeM · Hard ROxf | mAP: 44.3 -> 58.4 | Improved |
| increase re-ranking length K from 128 to 1536 (ablation) | R-GeM+Ours · Hard ROxf | mAP: 50.0 -> 59.1 | Improved |
| increase re-ranking length K from 128 to 1536 (ablation) | R-GeM+Ours · Hard RPar | mAP: 64.3 -> 75.3 | Improved |
| increase K from 128 to 1536 (ablation) | R-GeM+Ours · re-ranking latency | latency (ms): 7 -> 55 | Hurt |
| increase K from 1024 to 1536 (ablation) | R-GeM+Ours · Hard ROxf | mAP: 58.4 -> 59.1 | No clear effect |
| shrink anchor list length L from 512 to 128 (ablation) | R-GeM+Ours · Hard ROxf | mAP: 58.4 -> 55.9 | Hurt |
| enlarge anchor list length L from 512 to 1280 (ablation) | R-GeM+Ours · Hard ROxf | mAP: 58.4 -> 56.6 | Hurt |
| remove self-attention (transformer depth 0) (ablation) | affinity-feature re-ranker · Hard ROxf | mAP | Hurt |
| increase MSE loss weight lambda up to 0.2 (ablation) | re-ranker loss · Hard ROxf | mAP | Improved |
| make lambda excessively large (ablation) | re-ranker loss · Hard ROxf | mAP | Hurt |
| add data augmentation with multi-backbone affinity features (AugrSfM120k) (ablation) | Ours · Hard ROxf | mAP: 58.4 -> 60.0 | Improved |
| add data augmentation (AugrSfM120k) (ablation) | Ours · Hard RPar | mAP: 74.4 -> 76.3 | Improved |
Closest papers by meaning
- Re-ranking for image retrieval and transductive few-shot classification (NeurIPS 2021 · similarity 0.91)
- Instance-Level Image Retrieval Using Reranking Transformers (ICCV 2021 · similarity 0.89)
- Beyond Instance-Level Image Retrieval: Leveraging Captions to Learn a Global Visual Representation for Semantic Retrieval (CVPR 2017 · similarity 0.89)
- LOCORE: Image Re-ranking with Long-Context Sequence Modeling (CVPR 2025 · similarity 0.87)
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.87)
- QuARI: Query Adaptive Retrieval Improvement (NeurIPS 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.
