Data directories are missing for training Moco, and supplemental materials.

[halfjoe]

Dear Authors,

First of all I want to thank you for presenting this great work along with public codes. It really inspires the following work. And I have encountered some small problems during running the codes myself. It would be a great help if you could offer some tips about them.

First, while trying to re-train Moco, it reports that the data folder 'moco/ffhq_deg_q' and 'moco/ffhq_deg_k' are missing. Could you please provide some instructions on how these folders are generated? In the paper it looks that the paired positive data are generated during training for each iteration, yet in the codes it seems to be pre-generated.

Second, the paper mentions there are supplemental materials, and could you please leave a link for downloading them?

Thanks for your time, and looking forward to your reply!

Best Regards,
Fangzhou

[halfjoe]

Plus, in the class Panini_MFR, requires_grad_() are set to be True for self.generator (line 302) and self.deg_encoder (line 332). But according to the paper, these two models should be fixed during training other parts. Could you please provide some explanation over this?
Thanks!