This release is mainly for storing pre-trained models in VQFR v2.
The following are the changes comparing to VQFR v1:
- Reorganize the codebase and clean most parts of the research codes.
- Emphasize restoration quality and balance fidelity with user control.
- Reduce the model size and speed up its training and inference.