why the inception scores and FIDs are not comparable with the TF version?

ajbrock / BigGAN-PyTorch

The author's officially unofficial PyTorch BigGAN implementation.

MIT License

2.84k stars 470 forks source link

why the inception scores and FIDs are not comparable with the TF version? #83

Open KK666-AI opened 3 years ago

KK666-AI commented 3 years ago

Dear author, I notice that you say the computed inception scores and FIDs in the inception_utils.py are not comparable with the original TF IS/FID code. what's the mainly difference between these two implementations?

Best, Ken