Research map / Label-efficient and robust learning / Recognition and long-tail
ForestHash: Semantic Hashing With Shallow Random Forests and Tiny Convolutional Networks
Qiang Qiu, Jose Lezama, Alex Bronstein, Guillermo Sapiro
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 unsupervised information-theoretic code aggregation instead of random aggregation (ablation) | ForestHash · CIFAR-10 retrieval (reduced training, 48-bit, radius 0) | precision (%): 31.37 -> 34.02 | Improved |
| add supervised information-theoretic aggregation instead of random aggregation (ablation) | ForestHash · CIFAR-10 retrieval (reduced training, 48-bit, radius 0) | precision (%): 31.37 -> 33.86 | Improved |
| add semi-supervised aggregation instead of random aggregation (ablation) | ForestHash · CIFAR-10 retrieval (reduced training, 48-bit, radius 0) | precision (%): 31.37 -> 34.05 | Improved |
| add semi-supervised aggregation instead of random aggregation (ablation) | ForestHash · CIFAR-10 retrieval (reduced training, 48-bit, radius 0) | recall (%): 2.74 -> 4.12 | Improved |
| add semi-supervised instead of unsupervised aggregation (ablation) | ForestHash · CIFAR-10 retrieval (reduced training, 48-bit, radius <= 2) | precision (%): 34.55 -> 33.73 | Mixed |
| add low-rank loss instead of softmax loss only (ablation) | ForestHash CNN4 · CIFAR-10 retrieval (reduced training, 48-bit, radius 0) | precision (%): 22.72 -> 28.66 | Improved |
| add low-rank loss instead of softmax loss only (ablation) | ForestHash CNN2 · CIFAR-10 retrieval (reduced training, 48-bit, radius 0) | precision (%): 23.00 -> 29.30 | Improved |
| use more complex CNN4 learner instead of CNN2 under reduced training (ablation) | ForestHash · CIFAR-10 retrieval (reduced training, 48-bit, radius 0) | precision (%): 29.30 -> 28.66 | No clear effect |
| use CNN4 instead of CNN2 learner with full training set (ablation) | ForestHash · CIFAR-10 retrieval (50k train, official split), 48-bit | mAP: 0.80 -> 0.84 | Improved |
| use VGG16 instead of CNN4 learner with full training set (ablation) | ForestHash · CIFAR-10 retrieval (50k train, official split), 48-bit | mAP: 0.84 -> 0.89 | Improved |
| embed light-weight CNN2 weak learner instead of RBF kernel learner (ablation) | ForestHash · MNIST retrieval (36-bit, radius 0), 6,000 samples per class | precision (%): 86.53 -> 97.99 | Improved |
| embed light-weight CNN2 weak learner instead of RBF kernel learner (ablation) | ForestHash · MNIST retrieval (36-bit, radius 0), 30 samples per class | precision (%): 79.38 -> 89.56 | Improved |
Closest papers by meaning
- Deep Supervised Hashing for Fast Image Retrieval (CVPR 2016 · similarity 0.89)
- Supervised Learning of Semantics-Preserving Hash via Deep Convolutional Neural Networks (TPAMI 2017 · similarity 0.89)
- From Hashing to CNNs: Training Binary Weight Networks via Hashing (AAAI 2018 · similarity 0.89)
- Discretely Coding Semantic Rank Orders for Supervised Image Hashing (CVPR 2017 · similarity 0.88)
- KALAHash: Knowledge-Anchored Low-Resource Adaptation for Deep Hashing (AAAI 2025 · similarity 0.87)
- Deep Unsupervised Hashing via External Guidance (ICML 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.
