Research map / Detection and segmentation / CNN features
Deep Region Hashing for Generic Instance Search from Images
Jingkuan Song, Tao He, Lianli Gao, Xing Xu, Heng Tao Shen
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 |
|---|---|---|---|
| increase hash code length from 128 to 4096 bits (ablation) | gDRH · Oxford 5K | mAP: 0.425 -> 0.783 | Improved |
| increase hash code length from 1024 to 4096 bits (ablation) | gDRH · Oxford 5K | mAP: 0.748 -> 0.783 | No clear effect |
| replace real-valued max conv5 features with 4096-bit hash codes | gDRH · Oxford 5K | mAP: 0.797 -> 0.783 | Hurt |
| use RPN region proposals instead of sliding window (ablation) | DRH · Oxford 5K | mAP: 0.783 -> 0.780 | Hurt |
| vary sliding window overlap lambda (ablation) | DRH · Oxford 5K | mAP: 0.781 -> 0.783 | No clear effect |
| increase top M re-ranked images from 100 to 800 (ablation) | DRH · Oxford 5K | mAP: 0.776 -> 0.786 | No clear effect |
| add local region hashing re-ranking (lDRH) (ablation) | gDRH · Oxford 5K | mAP: 0.745 -> 0.783 | Improved |
| add global query expansion gQE with q=6 (ablation) | gDRH · Oxford 5K | mAP: 0.745 -> 0.815 | Improved |
| combine gQE and lQE with lDRH (ablation) | DRH · Oxford 5K | mAP: 0.745 -> 0.851 | Improved |
| use 512-bit hash codes instead of real-valued conv5 features for search | Tolias et al. conv5 search · Oxford 105K | query time (ms): 1078 -> 3 | Improved |
Closest papers by meaning
- Deep Supervised Hashing for Fast Image Retrieval (CVPR 2016 · similarity 0.87)
- On Trivial Solution and High Correlation Problems in Deep Supervised Hashing (AAAI 2018 · similarity 0.86)
- Deep Image Retrieval: Learning Global Representations for Image Search (ECCV 2016 · similarity 0.86)
- Deep Hashing Network for Efficient Similarity Retrieval (AAAI 2016 · similarity 0.86)
- Simultaneous Feature Aggregating and Hashing for Large-Scale Image Search (CVPR 2017 · similarity 0.85)
- Deep Unsupervised Hashing via External Guidance (ICML 2025 · similarity 0.84)
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.
