Research map / Learning theory and algorithms / Prediction and decision losses
Local Orthogonal-Group Testing
Ahmet Iscen, Ondrej Chum
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 |
|---|---|---|---|
| learn decoder U from PQ-compressed memory vectors instead of uncompressed (ablation) | Ours + PQ (c=16) · Oxford105k | mAP: 44.8 -> 62.3 | Improved |
| learn decoder U from PQ-compressed memory vectors instead of uncompressed (ablation) | Ours + PQ (c=32) · Oxford105k | mAP: 52.2 -> 70.5 | Improved |
| learn decoder U from PQ-compressed memory vectors instead of uncompressed (ablation) | Ours + PQ (c=256) · Oxford105k | mAP: 77.7 -> 79.2 | Improved |
| orthogonal grouping without decoder correction (LO-GT vs RND) (ablation) | 0-order decoder · Oxford105k | mAP | No clear effect |
| add decoder correction to orthogonal grouping (LO-GT*) (ablation) | LO-GT, 0-order decoder · Oxford105k | mAP | Improved |
| add decoder correction to random grouping (RND*) (ablation) | RND, 0-order decoder · Oxford105k | mAP | Hurt |
| use 1-order decoder instead of 0-order (ablation) | LO-GT* · Oxford105k | mAP | Improved |
| sparse decoder with L=300 non-zeros per column (ablation) | LO-GT*, 1-order decoder · Oxford105k | complexity ratio: 0.36 -> 0.23 | Improved |
| add cascade decoder (U0 + U1) (ablation) | sparse 1-order decoder, L=300 · Oxford105k | complexity ratio | Improved |
| increase decoder complexity (correction benefit) (ablation) | LO-GT* · Oxford105k | mAP | Mixed |
| split data into more batches for indexing | Ours (LO-GT*) · Oxford1M | mAP | No clear effect |
| split data into more batches for indexing | DL (dictionary learning) · Oxford1M | mAP | Hurt |
Closest papers by meaning
- Efficient Large-Scale Similarity Search Using Matrix Factorization (CVPR 2016 · similarity 0.89)
- Practical Near Neighbor Search via Group Testing (NeurIPS 2021 · similarity 0.87)
- Approximate Nearest Neighbor Search with Window Filters (ICML 2024 · similarity 0.86)
- Group Testing for Accurate and Efficient Range-Based Near Neighbor Search for Plagiarism Detection (ECCV 2024 · similarity 0.86)
- SOAR: Improved Indexing for Approximate Nearest Neighbor Search (NeurIPS 2023 · similarity 0.86)
- Composite Quantization (TPAMI 2018 · 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.
