Research map / Image generation / Autoregressive and latent generation
Deep Supervised Hashing for Fast Image Retrieval
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 |
|---|---|---|---|
| learn CNN image representations jointly with similarity-preserving binary codes | DSH versus the strongest listed baseline DNNH · CIFAR-10 | mAP (48-bit): 0.5904 -> 0.6755 | Helped |
| learn CNN image representations jointly with similarity-preserving binary codes | DSH versus the strongest listed baseline DNNH · NUS-WIDE | mAP (48-bit): 0.5248 -> 0.5621 | Helped |
| add an L1 regularizer that pushes real-valued outputs toward +1/-1 (ablation) | DSH · CIFAR-10 | mAP (12-bit): 0.5497 -> 0.6157 | Helped |
| add an L1 regularizer that pushes real-valued outputs toward +1/-1 (ablation) | DSH · NUS-WIDE | mAP (12-bit): 0.5076 -> 0.5483 | Helped |
| increase regularizer weight from α=0.01 to α=0.1 (ablation) | DSH · CIFAR-10 | mAP (12-bit): 0.6157 -> 0.4337 | Hurt |
| increase regularizer weight from α=0.01 to α=0.1 (ablation) | DSH · NUS-WIDE | mAP (12-bit): 0.5483 -> 0.4493 | Hurt |
| replace the output regularizer with sigmoid relaxation (ablation) | DSH · CIFAR-10 and NUS-WIDE | mAP (12-bit): CIFAR-10: 0.6157 vs 0.1451, 0.2812, 0.4788, 0.2196; NUS-WIDE: 0.5483 vs 0.4876, 0.5067, 0.4838, 0.4638 | Hurt |
| train larger-code models from scratch rather than finetuning the 12-bit model (ablation) | DSH, longer code lengths · CIFAR-10 and NUS-WIDE | mAP: CIFAR-10 24/36/48-bit: 0.6512/0.6607/0.6755 finetuned vs 0.6524/0.6433/0.6213 scratch; NUS-WIDE: 0.5513/0.5582/0.5621 vs | Mixed |
| generate pairwise training examples online from mini-batches instead of using offline Siamese pairs (ablation) | DSH training · CIFAR-10 and NUS-WIDE | training loss | Helped |
| increase code length from 12 to 48 bits using finetuning (ablation) | DSH · CIFAR-10 and NUS-WIDE | mAP: CIFAR-10: 0.6157 -> 0.6755; NUS-WIDE: 0.5483 -> 0.5621 | Helped |
Closest papers by meaning
- Deep Hashing Network for Efficient Similarity Retrieval (AAAI 2016 · similarity 0.95)
- Deep Quantization Network for Efficient Image Retrieval (AAAI 2016 · similarity 0.93)
- Deep Supervised Discrete Hashing (NeurIPS 2017 · similarity 0.92)
- HashGAN: Deep Learning to Hash With Pair Conditional Wasserstein GAN (CVPR 2018 · similarity 0.92)
- Pairwise-Label-Based Deep Incremental Hashing with Simultaneous Code Expansion (AAAI 2024 · similarity 0.91)
- Deep Graph Online Hashing for Multi-Label Image Retrieval (AAAI 2025 · similarity 0.89)
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.