Research map / Label-efficient and robust learning / Semi-supervised learning
Semi-Supervised Generative Adversarial Hashing for Image Retrieval
Guan'an Wang, Qinghao Hu, Jian Cheng, Zengguang Hou
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 adversary ranking loss (with generative model) to supervised-only baseline (ablation) | deep hashing model · CIFAR-10 | mAP: improves the baseline by 3.2% ~4.0% | Improved |
| add adversary ranking loss (with generative model) to supervised-only baseline (ablation) | deep hashing model · NUS-WIDE | mAP: improves the baseline by 0.9% ~2.8% | Improved |
| add semi-supervised ranking loss to supervised-only baseline (ablation) | deep hashing model · CIFAR-10 | mAP: improves the baseline by 4.8% ~5.7% | Improved |
| add semi-supervised ranking loss to supervised-only baseline (ablation) | deep hashing model · NUS-WIDE | mAP: improves the baseline by 3.2% ~5.1% | Improved |
| add adversary ranking loss on top of semi-supervised ranking loss (w/ ssr -> SSGAH) (ablation) | SSGAH · CIFAR-10 (48bits) | mAP: 0.846 -> 0.855 | Improved |
| add semi-supervised ranking loss on top of adversary ranking loss (w/ ar -> SSGAH) (ablation) | SSGAH · NUS-WIDE (48bits) | mAP: 0.812 -> 0.865 | Improved |
| add adversary ranking loss on top of semi-supervised ranking loss (ablation) | SSGAH · NUS-WIDE (12bits) | mAP: 0.810 -> 0.835 | Improved |
| use semi-supervised generative adversarial hashing instead of supervised-only baseline with less supervision | deep hashing model · CIFAR-10 and NUS-WIDE (48bits) | mAP | Improved |
| use unlabeled data via semi-supervised deep hashing instead of supervised deep hashing (NINH, CNNH) | deep hashing · CIFAR-10 and NUS-WIDE | mAP | Improved |
| use CNN fc7 features instead of hand-crafted features for traditional hashing | SDH · CIFAR-10 (48bits) | mAP: 0.360 -> 0.542 | Improved |
| use CNN fc7 features instead of hand-crafted features for traditional hashing | ITQ · NUS-WIDE (48bits) | mAP: 0.477 -> 0.661 | Improved |
| use SSGAH instead of SSDH for retrieval of unseen classes | hashing · NUS-WIDE unseen classes (48bits) | mAP: 0.551 -> 0.579 | Improved |
Closest papers by meaning
- SSAH: Semi-Supervised Adversarial Deep Hashing with Self-Paced Hard Sample Generation (AAAI 2020 · similarity 0.94)
- Prototype-Supervised Adversarial Network for Targeted Attack of Deep Hashing (CVPR 2021 · similarity 0.90)
- Unsupervised Deep Generative Adversarial Hashing Network (CVPR 2018 · similarity 0.90)
- Deep Supervised Adversarial Robust Hashing for Retrieval (TPAMI 2026 · similarity 0.88)
- AutoSSVH: Exploring Automated Frame Sampling for Efficient Self-Supervised Video Hashing (CVPR 2025 · similarity 0.86)
- Generalized Debiased Semi-Supervised Hashing for Large-Scale Image Retrieval (AAAI 2025 · 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.
