Research map / Image generation / GANs
FreGAN: Exploiting Frequency Components for Training GANs under Limited Data
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 |
|---|---|---|---|
| remove high-frequency discriminator (HFD) from FreGAN (ablation) | FreGAN · Pokemon 1024 | FID: 41.75 -> 38.88 | Helped |
| add HFA (remove w/o HFA) (ablation) | FreGAN · Pokemon 1024 | FID: 40.27 -> 38.88 | Helped |
| add frequency skip connection (FSC) (ablation) | FreGAN · Pokemon 1024 | FID: 39.41 -> 38.88 | Helped |
| add high-frequency discriminator (HFD) (ablation) | FreGAN · 100-shot-Obama | KID (x10^-3): 7.78 -> 3.76 | Helped |
| add HFD, HFA and FSC to FastGAN backbone | FastGAN · Flat 512 (36 imgs) | FID: 216.27 -> 178.10 | Helped |
| add FreGAN components to FastGAN | FastGAN · Skulls 1024 | FID: 101.94 -> 86.12 | Helped |
| add FreGAN components to FastGAN with more data | FastGAN · AFHQ-Cat | FID: 10.17 -> 6.62 | Helped |
| add FreGAN to FastGAN | FastGAN · Grumpy_cat / MetFace | KID: Grumpy_cat 3.41 -> 3.89; MetFace FastGAN 7.08 -> 5.93 | Mixed |
| apply HFD and HFA on multi-scale features (8,16,32) instead of single scale (ablation) | FreGAN · AnimalFace-Dog | FID: 51.86 -> 47.85 | Helped |
| fuse LH, HL and HH components instead of single component (ablation) | FreGAN HFA · AnimalFace-Dog | FID: 51.60 -> 47.85 | Helped |
| add FreGAN to MoCA attention-based GAN | MoCA · 100-shot-Obama | FID: 42.26 -> 34.13 | Helped |
| add FreGAN to Lecam regularization | Lecam · AnimalFace-Dog | FID: 54.88 -> 48.29 | Helped |
Closest papers by meaning
- On the Frequency Bias of Generative Models (NeurIPS 2021 · similarity 0.91)
- Spatial Frequency Bias in Convolutional Generative Adversarial Networks (AAAI 2022 · similarity 0.88)
- NICE: NoIse-modulated Consistency rEgularization for Data-Efficient GANs (NeurIPS 2023 · similarity 0.88)
- Frequency-Aware Deepfake Detection: Improving Generalizability through Frequency Space Domain Learning (AAAI 2024 · similarity 0.88)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · similarity 0.87)
- MS$^3$D: A RG Flow-Based Regularization for GAN Training with Limited Data (ICML 2024 · similarity 0.86)
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.