search

GaParmar / clean-fid

PyTorch - FID calculation with proper image resizing and quantization steps [CVPR 2022]
https://www.cs.cmu.edu/~clean-fid/
MIT License
971 stars 74 forks source link

A possible bug #7

Closed hangg7 closed 3 years ago

hangg7 commented 3 years ago

Hi Gaurav, thanks for sharing this amazing tool! I spot a block of suspicious lines that might worth your attention. Specifically, this resizing function of the default "clean" resizer: https://github.com/GaParmar/clean-fid/blob/d2a10b1f4f44e79ea08717a10702fb1c674b1830/cleanfid/resize.py#L43-L52

It seems to me that the output_size in L47 (s1, s2) supposes to be a (w, h) tuple whilie L48 expects it as (h, w). What do you think? This might mean that the default resizer only works for square output resolution.

Shouldn't be a big problem since it does not affect default behavior.

GaParmar commented 3 years ago

Hi Hang,

Thank you for pointing this out, I will address this in the next release and add more documentation here. However, this should not affect the FID behavior as the images are always resized to square output resolution 299x299.

Regards, Gaurav

hangg7 commented 3 years ago

Thanks! Closing this now.