Research map / Learning theory and algorithms / Prediction and decision losses
Fast Spectral Ranking for Similarity Search
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 online conjugate-gradient manifold ranking with the FSR spectral embedding | Regional image retrieval · Instre, Oxford5k, Paris6k | search time: 150 times faster on Oxford5k and Paris6k and 300 times faster on Instre | Helped |
| use the weighted variant, combining manifold and Euclidean scores | FSR approx on Oxford105k · Oxford105k | mAP: 90.2 -> 94.2 | Helped |
| add 100k distractors without truncating the dataset | Manifold retrieval · Oxford105k | mAP: +3% | Helped |
| reduce the spectral rank (ablation) | FSR rank-r · Paris6k | mAP | No clear effect |
| sparsify embeddings by retaining their largest entries (ablation) | Weighted FSR on Oxford105k · Oxford105k | mAP: 90% memory savings with only 2% drop of mAP | Hurt |
| compress descriptors with 256-byte product-quantization codes (ablation) | Weighted FSR on Oxford105k · Oxford105k | mAP: 94.4 -> 94.2 | Hurt |
| compress descriptors with 64-byte product-quantization codes (ablation) | Weighted FSR on Oxford105k · Oxford105k | mAP: 94.4 -> 91.1 | Hurt |
| build the graph using a smaller fraction of Instre images (ablation) | FSR global search · Instre | mAP: 80.5, 80.1, 78.3, 75.8, 70.2 and 40.6 | Mixed |
| use regional rather than global descriptors | Image retrieval · Instre, Oxford5k, Paris6k | mAP | Helped |
| apply 64-byte rather than 256-byte PQ compression (ablation) | Weighted FSR on Oxford105k · Oxford105k | mAP: 94.2 -> 91.1 | Hurt |
Closest papers by meaning
- Guided Similarity Separation for Image Retrieval (NeurIPS 2019 · similarity 0.89)
- Ensemble Deep Manifold Similarity Learning Using Hard Proxies (CVPR 2019 · similarity 0.87)
- Euclidean Embedding of Data Using Local Distances (NeurIPS 2026 · similarity 0.86)
- Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval (ICLR 2025 · similarity 0.84)
- Stabilizing Sample Similarity in Representation via Mitigating Random Consistency (ICML 2025 · similarity 0.84)
- Locality Preserving Markovian Transition for Instance Retrieval (ICML 2025 · similarity 0.83)
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.