Junhwa Song

@ckkelvinchan Thanks for the feedback. It seems good to me too. I will comment again based on your feedback.

@ckkelvinchan What should I do about this?

Of course. I think it is not difficult. If you need, I'll start writing a use case and unit tests for FID and KID metrics this week. However, as you...

@LeoXing1996 1. The bootstrapping logic of the current FID implementation came from the experimental setting of the paper **DEMYSTIFYING MMD GANS**. See the experiments section of that [paper](https://arxiv.org/abs/1801.01401). Generally, FID...

I'm so sorry for being too late. I was busy last week. @LeoXing1996 To clarify the scope of this PR, I want to remove bootstrapping logic of the current FID...

I want to work on this PR, but it's getting slower due to other schedules. I'll keep working on it and reply to you at least by this Sunday. @LeoXing1996...

Finally, I've just started work on this PR again. There are some questions. I think I can ensure test coverage for the InceptionV3, FID, and KID modules. However, I don't...

@LeoXing1996 Based on dummy input `np.random.randint(0, 256, (100, 224, 224, 3))` stored on disk, I compared the results between this implementation and [StyleGAN](https://github.com/NVlabs/stylegan3)'s implementation. The following table seems reasonable. |...

First, I apologize for not generating fake images from noise in an on-the-fly manner to reduce computation cost and time, unlike [StyleGAN](https://github.com/NVlabs/stylegan3) repo. However, in my view, how to generate...

@LeoXing1996 Great! I appreciate your hard work. Based on your suggestion, we will verify that the logic is working well and update the documentation.