CVPR2023
Robust Unsupervised StyleGAN Image Restoration
Yohan Poirier-Ginter, Jean-François Lalonde
摘要
https://lvsn.github.io/RobustUnsupervised/ Figure 1. Our unsupervised StyleGAN image restoration method is robust and effective across a variety of tasks (left, top to bottom: upsampling, denoising, deartifacting, and inpainting) and a wide range of degradation levels. Since it avoids the need for task-specific hyperparameter tuning, it can directly handle combinations of such degradations, e.g., pairs (center) or even all four (right).