pfnet-research/sngan_projection

Stat file for ImageNet dataset

Opened this issue · 1 comments

jamessealesmith commented

Hello,
We have been using your code and getting great results! Can I confirm that evaluation.calc_FID is not compatible with the results from evaluations/calc_ref_stats? I am trying to log FID scores when training on the ImageNet Dataset. If this assumption is correct, I will use evaluations/calc_intra_FID on model snapshots after training is complete. Thank you in advance for the response!

takerum commented

Hi,

You are right, and calc_FID in evaluation.py is no longer used in this repository. (this function is originally copied from https://github.com/pfnet-research/chainer-gan-lib)
Here is the link to the zip file including the stats files we used for calc_intra_FID in our experiments: https://www.dropbox.com/s/oqhc8frcjxkhuwz/ref_stats_tf.zip?dl=0.
Please use it if you will calculate the intra FID scores on ImageNet.

  • 👍1