fix bug when using --gfpgan-models-path

[avantcontra]

Description

• a simple description of what you're trying to accomplish

  • First, when using --gfpgan-models-path, for example, --gfpgan-models-path D:/custom/webui_models/GFPGAN, and selecting GFPGan as the Face restoration model in Restore faces, the detection_Resnet50_Final.pth file is still being downloaded to the original models\GFPGAN directory. This is because the save_dir parameter in facexlib.detection.load_file_from_url and facexlib.parsing.load_file_from_url is assigned as model_path instead of using the cmd parameter, which is user_path.

  • Second, in modelloader.load_models(), the ext_filter parameter is assigned as ext_filter="GFPGAN" (it should be ext_filter=['.pth']), which also causes the validation fail, resulting in a remote download every time.

  • Third, after fixing the above two issues, if I initially place the GFPGANv1.4.pth, detection_Resnet50_Final.pth, and parsing_parsenet.pth files in the --gfpgan-models-path, it will cause a new problem with gfpgan:

    File "C:\Users\Contra\anaconda3\envs\sd-pytorch2\lib\site-packages\gfpgan-1.3.8-py3.10.egg\gfpgan\utils.py", line 97, in init self.gfpgan.load_state_dict(loadnet[keyname], strict=True)
    KeyError: 'params'

    This is because detection_Resnet50_Final.pth or parsing_parsenet.pth is being mistakenly loaded as GFPGANv1.4.pth.

• a summary of changes in code

  • Add model_file_path. Because actually the parameter accepted by gfpgan.utils.load_file_from_url is the full path of the model file, not just its directory path.
  • Change ext_filter="GFPGAN" to ext_filter=['.pth'].
  • Add facexlib_path for save_dir of facexlib.detection.load_file_from_url and facexlib.parsing.load_file_from_url. The value depends on whether --gfpgan-models-path exists.
  • Add if 'GFPGAN' in os.path.basename(item) to take the GFPGANv1.4.pth to model_file.

• which issues it fixes, if any

Screenshots/videos:

Checklist:

• I have read contributing wiki page
• I have performed a self-review of my own code
• My code follows the style guidelines
• My code passes tests