search

devzhk / Implicit-Competitive-Regularization

Code for: Implicit Competitive Regularization in GANs
Apache License 2.0
114 stars 18 forks source link

cifar10.py #16

Closed zr-bee closed 1 year ago

zr-bee commented 1 year ago

Is the tflib.fid in cifar10.py a custom library, is it missing?

devzhk commented 1 year ago

Hey, The metrics directory contains files including cifar10.py are designed to compute FID (Fréchet Inception Distance) and IS (Inception Score) metrics. As of 2023, there are much better options for computing these metrics. If you are using PyTorch, I would suggest using the clean-fid package. On the other hand, if you are a TensorFlow user, the evaluation code available in the ADM repo from OpenAI's Guided Diffusion may be more suitable.

zr-bee commented 1 year ago

hey，Thanking your reply. I have other problem:  evalpt-cifar10-dc.sh，Where checkpoints/cifar-ACGD/ACGD-dc0.010 is obtained, it looks a bit mixed.Or my own level is relatively low

------------------ 原始邮件 ------------------ 发件人: "devzhk/Implicit-Competitive-Regularization" @.>; 发送时间: 2023年4月4日(星期二) 凌晨0:41 @.>; @.**@.>; 主题: Re: [devzhk/Implicit-Competitive-Regularization] cifar10.py (Issue #16)

Hey, The metrics directory contains files including cifar10.py are designed to compute FID (Fréchet Inception Distance) and IS (Inception Score) metrics. As of 2023, there are much better options for computing these metrics. If you are using PyTorch, I would suggest using the clean-fid package. On the other hand, if you are a TensorFlow user, the evaluation code available in the ADM repo from OpenAI's Guided Diffusion may be more suitable.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>