How to compute the IS using the tensorflow code from Du & Mordatch

wgrathwohl / JEM

Project site for "Your Classifier is Secretly an Energy-Based Model and You Should Treat it Like One"

Apache License 2.0

417 stars 63 forks source link

Closed sndnyang closed 4 years ago

sndnyang commented 4 years ago

Hi, thank you for your excellent work, the code and the pre-trained model.

I have a simple question. I want to reproduce the Inception Score in the paper. But the code is written in TensorFlow (https://github.com/openai/ebm_code_release/blob/master/test_inception.py), and your model is a PyTorch model.

So is it that converts the PyTorch model into a TensorFlow model with some libraries. Or are there any other methods to do it?

Thank you so much!

sndnyang commented 4 years ago

By the way, for FID score, the code compares two datasets. So it is

fid.py /path/to/generated_images /path/to/precalculated_stats.npz

right?

And how many images should I generate? 50000 (each class 5000)?