Research map / Label-efficient and robust learning / Recognition and long-tail
HashNet: Deep Learning to Hash by Continuation
Zhangjie Cao, Mingsheng Long, Jianmin Wang, Philip S. Yu
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 |
|---|---|---|---|
| add weighted maximum likelihood (ablation) | HashNet · ImageNet (64 bits) | MAP: 0.5992 -> 0.6835 | Improved |
| add weighted maximum likelihood (ablation) | HashNet · ImageNet (16 bits) | MAP: 0.3350 -> 0.5059 | Improved |
| add weighted maximum likelihood (ablation) | HashNet · NUS-WIDE (64 bits) | MAP: 0.6933 -> 0.7163 | Improved |
| add weighted maximum likelihood (ablation) | HashNet · MS COCO (64 bits) | MAP: 0.7348 -> 0.7362 | No clear effect |
| train with continuation to learn sgn codes instead of tanh plus binarization (ablation) | HashNet · ImageNet (64 bits) | MAP: 0.6061 -> 0.6835 | Improved |
| train with continuation to learn sgn codes instead of tanh plus binarization (ablation) | HashNet · NUS-WIDE (64 bits) | MAP: 0.7020 -> 0.7163 | Improved |
| train with continuation to learn sgn codes instead of tanh plus binarization (ablation) | HashNet · MS COCO (64 bits) | MAP: 0.7138 -> 0.7362 | Improved |
| use continuous similarity from label overlap (ablation) | HashNet · NUS-WIDE (64 bits) | MAP: 0.7163 -> 0.7259 | Improved |
| use continuous similarity from label overlap (ablation) | HashNet · ImageNet | MAP: 0.6835 -> 0.6835 | No clear effect |
| use HashNet instead of DHN | deep hashing · ImageNet (16 bits) | MAP: 0.3106 -> 0.5059 | Improved |
| use HashNet instead of DHN (zero-shot protocol) | deep hashing · ImageNet zero-shot (16 bits) | MAP: 0.2891 -> 0.4411 | Improved |
| use HashNet instead of DHN | deep hashing · MS COCO (64 bits) | MAP: 0.6944 -> 0.7362 | Improved |
Later papers that compared with this method
6 later accepted papers report HashNet as a baseline in their results tables; 5 of them report a better number on most of the results they share with it.
- Discretization Is Not Always Better: Rethinking Deep Quantization for Asymmetric Image Retrieval (AAAI 2026 · better on 0 of 1 shared result)
- LLC: Accurate, Multi-purpose Learnt Low-dimensional Binary Codes (NeurIPS 2021 · better on 5 of 5 shared results)
- Maximum-Margin Hamming Hashing (ICCV 2019 · better on 6 of 6 shared results)
- Hashing with Mutual Information (TPAMI 2019 · better on 4 of 4 shared results)
- Deep Cauchy Hashing for Hamming Space Retrieval (CVPR 2018 · better on 12 of 12 shared results)
- Transfer Adversarial Hashing for Hamming Space Retrieval (AAAI 2018 · better on 6 of 6 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- 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 Supervised Hashing for Fast Image Retrieval (CVPR 2016 · similarity 0.90)
- Deep Incremental Hashing Network for Efficient Image Retrieval (CVPR 2019 · similarity 0.90)
- HashGAN: Deep Learning to Hash With Pair Conditional Wasserstein GAN (CVPR 2018 · similarity 0.90)
- Pairwise-Label-Based Deep Incremental Hashing with Simultaneous Code Expansion (AAAI 2024 · similarity 0.88)
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.
