I have tested many face restoration models now (including CodeFormer, GFPGAN, RestoreFormer++, DMDNet, and GPEN) and GPEN has been the best-performing by far. IMO, it outpaces the options currently available (CodeFormer and GFPGAN) by a considerable margin.
Feature description
I have tested many face restoration models now (including CodeFormer, GFPGAN, RestoreFormer++, DMDNet, and GPEN) and GPEN has been the best-performing by far. IMO, it outpaces the options currently available (CodeFormer and GFPGAN) by a considerable margin.
GPEN repo can be found here: https://github.com/yangxy/GPEN
It also seems that FaceFusion have made the GPEN models available in ONNX form as well: https://github.com/facefusion/facefusion-assets
The roop-unleashed implementation of GPEN could also potentially be useful: https://github.com/C0untFloyd/roop-unleashed/blob/main/roop/processors/Enhance_GPEN.py
Thanks <3