CodeFormer is a Transformer-based prediction network used for image mosaic restoration. By learning discrete codebooks and decoders, it can reduce the uncertainty of restoration mappings and generate high-quality faces. It possesses excellent anti-degradation robustness and is suitable for synthetic datasets and real datasets.
Target Audience:
"Image mosaic removal"
Tool Features:
- Learning discrete codebooks and decoders
- Global facial composition modeling
- Controllable feature transformation
Tool Tags: Face restoration, discrete codebooks
