caiyuanhao1998/PNGAN
"Learning to Generate Realistic Noisy Images via Pixel-level Noise-aware Adversarial Training" (NeurIPS 2021)
Employs dual-domain alignment: image-space alignment via a pre-trained denoiser mapping fake and real noisy images to a noise-free solution space, combined with pixel-level adversarial training for noise distribution matching. The Simple Multi-scale Network (SMNet) generator captures realistic noise characteristics modeled as per-pixel random variables, enabling synthesis of training pairs without expensive ground-truth capture. Achieves SOTA results on SIDD, DND, PolyU, and Nam benchmarks when denoisers are trained on PNGAN-generated synthetic noisy images.
146 stars. No commits in the last 6 months.
Stars
146
Forks
19
Language
Python
License
MIT
Category
Last pushed
Jun 13, 2025
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/diffusion/caiyuanhao1998/PNGAN"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
zhangqianhui/Conditional-GAN
Tensorflow implementation for Conditional Convolutional Adversarial Networks.
mit-han-lab/data-efficient-gans
[NeurIPS 2020] Differentiable Augmentation for Data-Efficient GAN Training
kundan2510/pixelCNN
Theano reimplementation of pixelCNN architecture
EugenHotaj/pytorch-generative
Easy generative modeling in PyTorch
shaohua0116/DCGAN-Tensorflow
A Tensorflow implementation of Deep Convolutional Generative Adversarial Networks trained on...