Research map / Image generation / GANs
High-Fidelity Image Generation With Fewer Labels
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 |
|---|---|---|---|
| Use self-supervised and semi-supervised pre-training to infer conditioning labels, then add rotation prediction during GAN training | ImageNet 128×128 GAN generation with 20% labels · ImageNet 128×128, 20% labels | FID: 8.4 -> 7.7 | Helped |
| Add self-supervision during GAN training (ablation) | tranSS · ImageNet 128×128, 10% labels | FID: 8.9 -> 8.0 | Helped |
| Add self-supervision during GAN training (ablation) | tranSS · ImageNet 128×128, 5% labels | FID: 10.8 -> 10.4 | Helped |
| Add self-supervision during GAN training (ablation) | cotrainSS · ImageNet 128×128, 20% labels | FID: 13.9 -> 12.7 | Helped |
| Use hard predicted labels rather than soft predicted labels (ablation) | tranSS · ImageNet 128×128, 20% labels | FID: 10.4 -> 8.4 | Helped |
| Use hard predicted labels rather than soft predicted labels (ablation) | tranSS · ImageNet 128×128, 5% labels | FID: 15.4 -> 10.8 | Helped |
| Replace a single shared label with self-supervised representation clustering for conditioning | Unsupervised GAN generation · ImageNet 128×128, unsupervised | FID: 25.3 -> 23.2 | Helped |
| Use self-supervised representation clustering for conditioning instead of random labels | Unsupervised GAN generation · ImageNet 128×128, unsupervised | FID: 26.5 -> 23.2 | Helped |
| Add self-supervision during GAN training (ablation) | Clustering · ImageNet 128×128, unsupervised | FID: 23.2 -> 22.0 | Helped |
| Add self-supervision during GAN training (ablation) | Single label · ImageNet 128×128, unsupervised | FID: 25.3 -> 23.7 | Helped |
| Use more than 50 clusters for the clustering-based conditioning labels (ablation) | Clustering · ImageNet 128×128, unsupervised | FID | Hurt |
| Vary the discriminator rotation-loss weight over the tested values (ablation) | tranSSS · ImageNet 128×128 | FID | No clear effect |
Closest papers by meaning
- Large Scale GAN Training for High Fidelity Natural Image Synthesis (ICLR 2019 · similarity 0.91)
- CcGAN: Continuous Conditional Generative Adversarial Networks for Image Generation (ICLR 2021 · similarity 0.90)
- OSSGAN: Open-Set Semi-Supervised Image Generation (CVPR 2022 · similarity 0.90)
- Self-Supervised GANs via Auxiliary Rotation Loss (CVPR 2019 · similarity 0.90)
- Unlocking Pre-trained Image Backbones for Semantic Image Synthesis (CVPR 2024 · similarity 0.88)
- 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.