Research map / Detection and segmentation / CNN features
Instance-Level Image Retrieval Using Reranking Transformers
Fuwen Tan, Jiangbo Yuan, Vicente Ordonez
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 |
|---|---|---|---|
| replace geometric verification with RRT reranking | DELG R50 v2-clean top-100 · ROxf Hard | mAP: 57.9 -> 60.2 | Improved |
| replace geometric verification with RRT reranking | DELG R50 v2-clean top-100 · ROxf Medium | mAP: 78.3 -> 78.1 | No clear effect |
| rerank top-200 instead of top-100 with RRT (ablation) | RRT, DELG v1 · RPar Hard | mAP: 68.6 -> 71.6 | Improved |
| rerank top-400 instead of top-100 with RRT (ablation) | RRT, DELG v2-clean · ROxf Hard | mAP: 60.2 -> 64.2 | Improved |
| combine alphaQE with RRT reranking top-200 | alphaQE, DELG v2-clean · ROxf Hard | mAP: 54.6 -> 64.0 | Improved |
| replace ASMK rerank with RRT | DELG v2-clean · ROxf Hard | mAP: 49.3 -> 60.2 | Improved |
| replace SuperGlue (1024 descriptors) with RRT | DELG v2-clean · RPar Hard | mAP: 70.4 -> 75.1 | Improved |
| replace GV with RRT on GLDv2 retrieval | DELG v2-clean · GLDv2 private | mAP@100: 24.3 -> 27.0 | Improved |
| add RRT reranking on a frozen backbone | CO · SOP | R@1: 80.7 -> 81.8 | Improved |
| finetune the backbone jointly with RRT (ablation) | CO + RRT · SOP | R@1: 81.8 -> 84.5 | Improved |
| add position embedding to RRT (ablation) | RRT with SuperPoint · SOP | R@1: 50.2 -> 47.3 | Hurt |
| increase local descriptors from 500 to 1000 for RRT (ablation) | RRT, v2-clean · ROxf Hard | mAP: 60.2 -> 60.4 | No clear effect |
Closest papers by meaning
- Contextual Similarity Aggregation with Self-attention for Visual Re-ranking (NeurIPS 2021 · similarity 0.89)
- R2Former: Unified Retrieval and Reranking Transformer for Place Recognition (CVPR 2023 · similarity 0.89)
- LOCORE: Image Re-ranking with Long-Context Sequence Modeling (CVPR 2025 · similarity 0.88)
- QuARI: Query Adaptive Retrieval Improvement (NeurIPS 2025 · similarity 0.86)
- Instance-Level Composed Image Retrieval (NeurIPS 2025 · similarity 0.86)
- Cheb-GR: Rethinking K-nearest Neighbor Search in Re-ranking for Person Re-identification (CVPR 2025 · similarity 0.85)
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.
