Research map / Image generation / GANs
On Self Modulation for Generative Adversarial Networks
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 latent-vector-dependent modulation to generator batch normalization | GAN generators across unconditional settings · LSUN-bedroom, CelebA-HQ, CIFAR10, ImageNet | FID: 14.32 vs 17.10; 22.51 vs 26.15; 18.54 vs 20.08; 68.90 vs 78.62 | Helped |
| Add self-modulation while keeping hyperparameters fixed | GANs across paired settings · 144 paired comparisons | FID: 124/144 (86%) | Helped |
| Add self-modulation to ResNet generators | ResNet GANs · All considered datasets, losses, and regularization or normalization settings | FID: 4.3% to 33% | Helped |
| Add self-modulation to SNDCGAN generators | SNDCGAN · All considered datasets, losses, and regularization or normalization settings | FID: 87.5% | Mixed |
| Add self-modulation alongside label conditioning | G-Cond and P-cGAN · CIFAR10 and ImageNet | FID: CIFAR10: 21.08 -> 18.39; ImageNet: 80.43 -> 68.93; CIFAR10: 16.06 -> 14.19; ImageNet: 70.28 -> 66.09 | Helped |
| Add self-modulation to unconditional and conditional generators | ImageNet GANs trained for 500k steps · ImageNet unconditional and conditional | FID: 60.4 to 53.7; 50.6 to 43.9 | Helped |
| Add self-modulation to unconditional and conditional generators | ImageNet GANs trained for 500k steps · ImageNet unconditional and conditional | Inception Score: 14.1 to 15.1; 20.1 to 22.2 | Helped |
| Apply self-modulation to each generator batch-normalization layer instead of only one layer (ablation) | Self-Mod · Multiple datasets | FID | Helped |
| Concentrate self-modulation on later generator layers (ablation) | Self-Mod · Multiple datasets | FID | Helped |
| Use a more complex modulation architecture (ablation) | Self-Mod · Initial experiments | GAN performance | No clear effect |
Closest papers by meaning
- Self-supervised GAN: Analysis and Improvement with Multi-class Minimax Game (NeurIPS 2019 · similarity 0.88)
- Adversarial Self-Defense for Cycle-Consistent GANs (NeurIPS 2019 · similarity 0.88)
- Self-Supervised GANs with Label Augmentation (NeurIPS 2021 · similarity 0.88)
- Augmentation-Aware Self-Supervision for Data-Efficient GAN Training (NeurIPS 2023 · similarity 0.88)
- Top-k Training of GANs: Improving GAN Performance by Throwing Away Bad Samples (NeurIPS 2020 · similarity 0.87)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · 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.