Research map / Label-efficient and robust learning / Recognition and long-tail
Similarity Preserving Deep Asymmetric Quantization for Image Retrieval
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 |
|---|---|---|---|
| learn codebooks and binary codes for all database items (vs subset only, SPDAQ-LL) (ablation) | SPDAQ · CIFAR-10 24 bits | mAP (%): 17.1 -> 88.4 | Helped |
| learn codebooks and binary codes for all database items (vs subset only, SPDAQ-LL) (ablation) | SPDAQ · NUS-WIDE-21 24 bits | mAP (%): 49.6 -> 93.1 | Helped |
| add classification loss (vs SPDAQ-nC) (ablation) | SPDAQ · CIFAR-10 24 bits | mAP (%): 85.2 -> 88.4 | Helped |
| add classification loss (vs SPDAQ-nC) (ablation) | SPDAQ · NUS-WIDE-21 24 bits | mAP (%): 92.2 -> 93.1 | Helped |
| add classification loss (vs SPDAQ-nC) (ablation) | SPDAQ · CIFAR-10 8 bits | mAP (%): 78.6 -> 88.4 | Helped |
| quantization (vs unquantized SPDAQ-nQ, upper bound) (ablation) | SPDAQ · CIFAR-10 24 bits | mAP (%): 90.1 -> 88.4 | Hurt |
| discrete direct B learning (SPDAQ) vs Lagrangian form SPDAQ-LD (ablation) | SPDAQ · CIFAR-10 24 bits | mAP (%): 88.7 vs 88.4 | No clear effect |
| SPDAQ vs DVSQ* baseline | deep quantization · CIFAR-10 24 bits | mAP (%): 80.3 -> 88.4 | Helped |
| SPDAQ vs DTQ | deep quantization · MS-COCO 24 bits | mAP (%): 76.4 -> 84.5 | Helped |
| train on all database items with baselines (DPSH-ALL, DSDH-ALL) | deep hashing baselines · NUS-WIDE-21 training efficiency | mAP@5000: about 84% after more than 10 hours vs SPDAQ about 3.5 hours | No clear effect |
Closest papers by meaning
- Deep Visual-Semantic Quantization for Efficient Image Retrieval (CVPR 2017 · similarity 0.91)
- Discretization Is Not Always Better: Rethinking Deep Quantization for Asymmetric Image Retrieval (AAAI 2026 · similarity 0.90)
- Generalized Product Quantization Network for Semi-Supervised Image Retrieval (CVPR 2020 · similarity 0.89)
- Supervised Quantization for Similarity Search (CVPR 2016 · similarity 0.89)
- HiHPQ: Hierarchical Hyperbolic Product Quantization for Unsupervised Image Retrieval (AAAI 2024 · similarity 0.88)
- Pairwise-Label-Based Deep Incremental Hashing with Simultaneous Code Expansion (AAAI 2024 · 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.