Research map / Label-efficient and robust learning / Recognition and long-tail
Supervised Learning of Semantics-Preserving Hash via Deep Convolutional Neural Networks
Huei‐Fang Yang, Kevin Lin, Chu‐Song Chen
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 binarization term E2 (beta=1) to label-consistency loss (ablation) | SSDH · CIFAR-10 (48 bits) | mAP (%): 90.70 -> 91.19 | Improved |
| add equal-sparsity term E3 (gamma=1) on top of E1 and E2 (ablation) | SSDH · CIFAR-10 (48 bits) | mAP (%): 91.19 -> 91.45 | Improved |
| use L2 instead of L1 norm loss (ablation) | SSDH · CIFAR-10 (12 bits) | mAP (%): 87.25 -> 90.59 | Improved |
| use L2 instead of L1 norm loss (ablation) | SSDH · MNIST (12 bits) | mAP (%): 98.90 -> 99.31 | No clear effect |
| use L2 margin instead of L1 margin in maximum-margin loss (ablation) | SSDH · NUS-WIDE (12 bits) | mAP (%): 71.73 -> 85.17 | Improved |
| use L2 margin instead of L1 margin in maximum-margin loss (ablation) | SSDH · UT-ZAP50K (48 bits) | mAP (%): 65.94% -> 71.91% | Improved |
| replace two-stage fine-tuned AlexNet + CCA-ITQ with end-to-end SSDH | hashing on Yahoo-1M · Yahoo-1M (128 bits) | mAP (%): 61.69 -> 66.63 | Improved |
| replace AlexNet with VGG16 backbone | SSDH · ILSVRC2012 (512 bits) | mAP (%): 46.07 -> 61.47 | Improved |
| replace AlexNet with VGG16 backbone | SSDH · CIFAR-10 (48 bits) | mAP (%): 91.45 -> 92.69 | Improved |
| use VGG11 backbone instead of VGG16 | SSDH · CIFAR-10 (48 bits) | mAP (%): 92.69 -> 88.40 | Hurt |
| replace VGG16 fully connected layers with average pooling (VGG-Avg) | SSDH · CIFAR-10 (48 bits) | mAP (%): 92.69 -> 90.75 | Mixed |
| remove 50% of labels of multi-label training images | SSDH (VGG16) · NUS-WIDE (48 bits) | mAP (%): 88.97% -> 88.02% | No clear effect |
Closest papers by meaning
- Deep Supervised Hashing for Fast Image Retrieval (CVPR 2016 · similarity 0.92)
- Deep Hashing Network for Efficient Similarity Retrieval (AAAI 2016 · similarity 0.92)
- Supervised Deep Hashing for Hierarchical Labeled Data (AAAI 2018 · similarity 0.91)
- Deep Supervised Discrete Hashing (NeurIPS 2017 · similarity 0.91)
- SADIH: Semantic-Aware DIscrete Hashing (AAAI 2019 · similarity 0.90)
- Pairwise-Label-Based Deep Incremental Hashing with Simultaneous Code Expansion (AAAI 2024 · similarity 0.89)
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.
