Converting GFPGANv1 to GFPGANv1Clean #210

Open
opened 2026-01-29 21:45:42 +00:00 by claunia · 4 comments
Owner

Originally created by @bocsiboti on GitHub (Jul 19, 2022).

Is the convert_gfpganv_to_clean.py script supposed to work?

I successfully converted (with no error) the GFPGANv1.pth model to a clean version but its prediction are non-sense.

I tried to change the channel_multiplier, narrow, fix_decoder parameters of the model, but got the same results.

The (1) input image, (2) result of the v1 model prediction, and (3) result of the converted to cleaned predictions can be seen here: https://drive.google.com/drive/folders/1tIa6Uq1cMST4tOLt8EzGhOlnYKZLyhqL?usp=sharing

Originally created by @bocsiboti on GitHub (Jul 19, 2022). Is the convert_gfpganv_to_clean.py script supposed to work? I successfully converted (with no error) the GFPGANv1.pth model to a clean version but its prediction are non-sense. I tried to change the channel_multiplier, narrow, fix_decoder parameters of the model, but got the same results. The (1) input image, (2) result of the v1 model prediction, and (3) result of the converted to cleaned predictions can be seen here: https://drive.google.com/drive/folders/1tIa6Uq1cMST4tOLt8EzGhOlnYKZLyhqL?usp=sharing
Author
Owner

@elen07zz commented on GitHub (Jul 21, 2022):

.

@elen07zz commented on GitHub (Jul 21, 2022): .
Author
Owner

@magicse commented on GitHub (Aug 16, 2022):

Could you explain how You made conversion?

@magicse commented on GitHub (Aug 16, 2022): Could you explain how You made conversion?
Author
Owner

@ggsddu7 commented on GitHub (Jun 25, 2023):

convert_gfpganv_to_clean.py use to convert GFPGANBilinear ==> GFPGANv1Clean

@ggsddu7 commented on GitHub (Jun 25, 2023): convert_gfpganv_to_clean.py use to convert GFPGANBilinear ==> GFPGANv1Clean
Author
Owner

@shubhamghimire commented on GitHub (Jan 22, 2024):

@bocsiboti @elen07zz @magicse @Eyesfront-ZJG I have encountered the same issue. I am training the GFPGAN model with my own datasets. The images transformed using the trained model has no problem.
Then I converted the trained model to clean model using the convert_gfpganv_to_clean.py.
But the images transformed using the clean model becomes strange like there is red color painted all over the transformed image. The predictions of the clean version are non-sense.

Did you guys able to solve the problem?

@shubhamghimire commented on GitHub (Jan 22, 2024): @bocsiboti @elen07zz @magicse @Eyesfront-ZJG I have encountered the same issue. I am training the GFPGAN model with my own datasets. The images transformed using the trained model has no problem. Then I converted the trained model to clean model using the convert_gfpganv_to_clean.py. But the images transformed using the clean model becomes strange like there is red color painted all over the transformed image. The predictions of the clean version are non-sense. Did you guys able to solve the problem?
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: TencentARC/GFPGAN#210