Research map / Label-efficient and robust learning / Recognition and long-tail
A Unified Generative Adversarial Network Training via Self-Labeling and Self-Attention
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 self-labeling without self-attention (ablation) | BigGAN on CIFAR-10 unlabeled · CIFAR-10 unlabeled | FID-5ep: 7.45 ± 0.17 -> 7.74 ± 0.20 | Hurt |
| add self-labeling without self-attention (ablation) | BigGAN on CIFAR-10 unlabeled · CIFAR-10 unlabeled | mean FID: 6.96 ± 0.20 -> 7.00 ± 0.34 | No clear effect |
| add self-labeling and self-attention (ablation) | BigGAN on CIFAR-10 unlabeled · CIFAR-10 unlabeled | FID-5ep: 7.45 -> 7.28 | Helped |
| add self-labeling and self-attention (ablation) | BigGAN on CIFAR-10 unlabeled · CIFAR-10 unlabeled | mean FID: 6.96 -> 6.81 | Helped |
| use K=5 artificial classes instead of the baseline setting (ablation) | BigGAN on CIFAR-10 unlabeled · CIFAR-10 unlabeled | FID: 6.96 -> 5.97 | Helped |
| use K=2 artificial classes instead of the baseline setting (ablation) | BigGAN on CIFAR-10 unlabeled · CIFAR-10 unlabeled | FID: 6.96 -> 6.07 | Helped |
| increase artificial classes to K=50 (ablation) | BigGAN on CIFAR-10 unlabeled · CIFAR-10 unlabeled | FID: 6.96 -> 7.54 | Hurt |
| combine artificial labels with real labels instead of using real labels alone (ablation) | BigGAN on CIFAR-10 labeled · CIFAR-10 labeled | mean FID: 4.57 -> 4.35 | Helped |
| use self-labeling and self-attention | BigGAN on CIFAR-10 unlabeled · CIFAR-10 unlabeled | mean FID: 8.57 -> 6.81 | Helped |
| use self-labeling and self-attention | BigGAN on CIFAR-10 labeled · CIFAR-10 labeled | mean FID: 5.39 -> 4.35 | Helped |
| use self-labeling and self-attention | BigGAN on STL-10 unlabeled · STL-10 unlabeled | FID: 32.85 -> 30.91 | Helped |
| use self-labeling and self-attention | BigGAN on SVHN unlabeled · SVHN unlabeled | FID: 2.44 -> 2.19 | Helped |
Closest papers by meaning
- Self-Supervised GANs with Label Augmentation (NeurIPS 2021 · similarity 0.93)
- Self-supervised GAN: Analysis and Improvement with Multi-class Minimax Game (NeurIPS 2019 · similarity 0.90)
- Augmentation-Aware Self-Supervision for Data-Efficient GAN Training (NeurIPS 2023 · similarity 0.90)
- Quality-Aware Self-Training on Differentiable Synthesis of Rare Relational Data (AAAI 2023 · similarity 0.89)
- Improved Techniques for Training GANs (NeurIPS 2016 · similarity 0.89)
- Fusing Conditional Submodular GAN and Programmatic Weak Supervision (AAAI 2024 · similarity 0.88)
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.