Research map / Label-efficient and robust learning / Recognition and long-tail
Supervised Quantization for Similarity Search
Check 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 |
|---|---|---|---|
| use supervised quantization rather than SDH | CIFAR-10, 64-bit codes · CIFAR-10 | MAP: 23.66% improvement | Helped |
| use supervised quantization rather than SDH | NUS-WIDE, 16-bit codes · NUS-WIDE | MAP: 4.65% improvement | Helped |
| use supervised quantization rather than composite quantization | CIFAR-10, MNIST, and NUS-WIDE, 16-bit codes · CIFAR-10, MNIST, and NUS-WIDE | MAP: 42.57% improvement on CIFAR-10 with 16 bits; 46.14% on MNIST with 16 bits; 15.39% on NUS-WIDE with 16 bits | Helped |
| replace classification loss with triplet loss (ablation) | SQ · CIFAR-10 and MNIST | MAP: CIFAR-10: 0.3284 vs 0.6045 at 16 bits; MNIST: 0.4347 vs 0.9329 at 16 bits | Hurt |
| add learned feature transformation (ablation) | SQ · CIFAR-10 | MAP: 0.5140 -> 0.6045 at 16 bits; 0.5174 -> 0.6855 at 32 bits; 0.5274 -> 0.7042 at 64 bits; 0.5301 -> 0.7120 at 128 bits | Helped |
| add learned feature transformation (ablation) | SQ · MNIST | MAP: 0.4534 -> 0.9329 at 16 bits; 0.4538 -> 0.9374 at 32 bits; 0.4617 -> 0.9377 at 64 bits; 0.4650 -> 0.9400 at 128 bits | Helped |
| use supervised quantization rather than SDH | MNIST · MNIST | MAP | No clear effect |
| use supervised quantization rather than Euclidean distance on original CNN features | ImageNet · ImageNet | MAP | Helped |
| use classification loss instead of triplet loss (ablation) | SQ · CIFAR-10 and MNIST | MAP: CIFAR-10: 0.6045 vs 0.3284 at 16 bits; MNIST: 0.9329 vs 0.4347 at 16 bits | Helped |
Closest papers by meaning
- Collaborative Quantization for Cross-Modal Similarity Search (CVPR 2016 · similarity 0.89)
- Multiscale Quantization for Fast Similarity Search (NeurIPS 2017 · similarity 0.89)
- Similarity Preserving Deep Asymmetric Quantization for Image Retrieval (AAAI 2019 · similarity 0.89)
- Deep Visual-Semantic Quantization for Efficient Image Retrieval (CVPR 2017 · similarity 0.88)
- Weakly Supervised Deep Hyperspherical Quantization for Image Retrieval (AAAI 2021 · similarity 0.87)
- HiHPQ: Hierarchical Hyperbolic Product Quantization for Unsupervised Image Retrieval (AAAI 2024 · 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.