Research map / Image generation / Autoregressive and latent generation
Deep Incremental Hashing Network for Efficient 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 |
|---|---|---|---|
| Keep original database codes fixed and learn codes for incoming images incrementally | DIHN3+ADSH · CIFAR-10 and NUS-WIDE | MAP: CIFAR-10: 0.8898 -> 0.8916 (12 bits), 0.9390 -> 0.9456 (48 bits); NUS-WIDE: 0.8400 -> 0.8353 (12 bits), 0.9055 -> 0.9115 | Mixed |
| Use a 0.2 sample rate for original codes instead of the full set (ablation) | DIHN*+ADSH · CIFAR-10 and NUS-WIDE | MAP: CIFAR-10: 0.8916 -> 0.7823 (12 bits), 0.9456 -> 0.9079 (48 bits); NUS-WIDE: 0.8353 -> 0.8113 (12 bits), 0.9115 -> 0.9075 | Hurt |
| Use a 0.4 sample rate for original codes instead of the full set (ablation) | DIHN*+ADSH · CIFAR-10 and NUS-WIDE | MAP: CIFAR-10: 0.8916 -> 0.8539 (12 bits), 0.9456 -> 0.9319 (48 bits); NUS-WIDE: 0.8353 -> 0.8300 (12 bits), 0.9115 -> 0.9078 | Mixed |
| Use a 0.6 sample rate for original codes instead of the full set (ablation) | DIHN*+ADSH · CIFAR-10 and NUS-WIDE | MAP: CIFAR-10: 0.8916 -> 0.8910 (12 bits), 0.9456 -> 0.9372 (48 bits); NUS-WIDE: 0.8353 -> 0.8321 (12 bits), 0.9115 -> 0.9111 | Mixed |
| Use a 0.8 sample rate for original codes instead of the full set (ablation) | DIHN*+ADSH · CIFAR-10 and NUS-WIDE | MAP: CIFAR-10: 0.8916 -> 0.8815 (12 bits), 0.9456 -> 0.9303 (48 bits); NUS-WIDE: 0.8353 -> 0.8291 (12 bits), 0.9115 -> 0.9103 | Mixed |
| Use only 20% labeled incremental images and generate codes for unlabeled images with the CNN (ablation) | DIHN3-S+ADSH · CIFAR-10 and NUS-WIDE | MAP: CIFAR-10: 0.8916 -> 0.7594 (12 bits), 0.9456 -> 0.8774 (48 bits); NUS-WIDE: 0.8353 -> 0.8151 (12 bits), 0.9115 -> 0.8640 | Hurt |
| Use 80% labeled incremental images and generate codes for the remainder with the CNN (ablation) | DIHN3-S+ADSH · CIFAR-10 and NUS-WIDE | MAP: CIFAR-10: 0.8916 -> 0.8769 (12 bits), 0.9456 -> 0.9352 (48 bits); NUS-WIDE: 0.8353 -> 0.8171 (12 bits), 0.9115 -> 0.8901 | Hurt |
| Increase the number of labeled incremental images (ablation) | DIHN3-S+ADSH · CIFAR-10 and NUS-WIDE | MAP | Helped |
| Use a smaller subset of original codes to reduce training work (ablation) | DIHN*+ADSH · NUS-WIDE | training time: 40 minutes -> 20 minutes | Helped |
| Use DIHN incremental training rather than retraining over the full database | DIHN3+ADSH versus ADSH · NUS-WIDE, 12-bit codes | training time: 28 minutes; 83 minutes | Helped |
| Use DIHN3 instead of ADSH for one training iteration | DIHN3 versus ADSH · NUS-WIDE | training time per iteration: ADSH: 122.78, 178.97, 216.26, 296.41; DIHN3: 65.06, 76.83, 79.88, 86.49 | Helped |
Closest papers by meaning
- Pairwise-Label-Based Deep Incremental Hashing with Simultaneous Code Expansion (AAAI 2024 · similarity 0.92)
- Deep Supervised Hashing for Fast Image Retrieval (CVPR 2016 · similarity 0.92)
- Deep Hashing Network for Efficient Similarity Retrieval (AAAI 2016 · similarity 0.92)
- Deep Quantization Network for Efficient Image Retrieval (AAAI 2016 · similarity 0.90)
- Deep Graph Online Hashing for Multi-Label Image Retrieval (AAAI 2025 · similarity 0.90)
- Learning Attribute-Aware Hash Codes for Fine-Grained Image Retrieval via Query Optimization (ICML 2025 · similarity 0.87)
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.