Research map / Image generation / GANs
Binary Generative Adversarial Networks for Image Retrieval
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 content and adversarial losses to neighborhood loss (ablation) | BGAN · CIFAR-10, 24/32/48-bit codes | mAP: 0.487 -> 0.512; 0.511 -> 0.531; 0.543 -> 0.558 | Helped |
| Remove adversarial loss (ablation) | BGAN · CIFAR-10, 24/32/48-bit codes | mAP: 0.247 -> 0.512; 0.379 -> 0.531; 0.497 -> 0.558 | Hurt |
| Remove content loss (ablation) | BGAN · CIFAR-10, 24/32/48-bit codes | mAP: 0.472 -> 0.512; 0.503 -> 0.531; 0.534 -> 0.558 | Hurt |
| Remove neighborhood structure loss (ablation) | BGAN · CIFAR-10 | hash-code diversity | Hurt |
| Use direct binary optimization rather than two-step binarization (ablation) | BGAN · CIFAR-10, 24/32/48-bit codes | mAP: 2.5%; 3.2%; 2.3% | Helped |
| Use tanh continuation instead of two-step binarization (ablation) | BGAN · CIFAR-10, 24/32/48-bit codes | mAP: 3.0%; 4.6%; 2.8% | Helped |
| Use app continuation instead of two-step binarization (ablation) | BGAN · CIFAR-10, 24 bits | mAP: 0.512 -> 0.537 | Helped |
| Use app continuation rather than tanh continuation (ablation) | BGAN · CIFAR-10, 24 bits | mAP: 0.542 -> 0.537 | Hurt |
| Use BGAN rather than the strongest listed unsupervised baseline | image retrieval hashing · CIFAR-10, NUS-WIDE, and Flickr | mAP: Table 2 reports BGAN and baseline values across 12, 24, 32, and 48 bits. | Helped |
Closest papers by meaning
- BinGAN: Learning Compact Binary Descriptors with a Regularized GAN (NeurIPS 2018 · similarity 0.90)
- HashGAN: Deep Learning to Hash With Pair Conditional Wasserstein GAN (CVPR 2018 · similarity 0.90)
- Unsupervised Deep Generative Adversarial Hashing Network (CVPR 2018 · similarity 0.89)
- Deep Supervised Hashing for Fast Image Retrieval (CVPR 2016 · similarity 0.88)
- Large Scale Adversarial Representation Learning (NeurIPS 2019 · similarity 0.88)
- Fast Training of Triplet-Based Deep Binary Embedding Networks (CVPR 2016 · similarity 0.87)
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.