Skip to content

Latest commit

 

History

History
35 lines (28 loc) · 1.36 KB

tensorflow_fid.md

File metadata and controls

35 lines (28 loc) · 1.36 KB
Raw

CleanFID vs TensorFlow-FID (legacy)

TensorFlow-legacy-FID refers to the official implementation of FID found here code

Here we document the differences between CleanFID and TensorFlow-legacy-FID. Additionally, we outline the steps that can be used to verify that using the mode="legacy_tensorflow" flag is equivalent to the official TensorFlow implementation .

Getting sample images

First download zipped file containing 5000 images generated by a StyleGAN2 model and 5000 randomly sampled from the FFHQ dataset.

wget <to_be_hosted>.zip

Compare the FID scores

  • First compute the FID score using the CleanFID with

    from cleanfid import fid

score = fid.compare_folders("test_fake/", "test_real/", mode="legacy_tensorflow")

    The above gets a score of 8.972544787448271

  • Next compute the FID using the [code]((https://github.com/bioinf-jku/TTUR)

    pip install tensorflow-gpu==1.14
pip install imageio
cd test && git clone https://github.com/bioinf-jku/TTUR
python fid.py test_fake/ test_real/

    The above gets a score of 8.972518580709163

Debugging Notes

  • The official TTUR repository is very sensitive to the the tensorflow version.
  • You may need to tune and modify some shape operations
  • I tested with tensorflow 1.14 and python 3.6