Question related to data-free ECD

[felix-yuxiang]

Hi @Gsunshine,

Great work, thank you for releasing the code!

I was wondering if you plan to release the code for the data-free ECD method. In your paper, you report a 1-step FID of 4.38 and a 2-step FID of 2.77 on ImageNet 64x64. However, I haven’t been able to reproduce these results with the current codebase. Could you please provide some guidance or insights on how to achieve similar performance?