How did you split the dataset?

[Kihensarn]

It's a wonderful work to manipulate face by using only text. But when I try the method in mapper folder，I see that the number of training samples is 24176 and the number of test samples is 2824 in the train_face.pt and test_face.pt, respectively. I am confused that the total numbers is not 30000 because there are 30000 samples in CelebA-HQ dataset. And could you also let me know the corresponding filename in the train_face.pt and test_face.pt？

[orpatashnik]

Hi @Kihensarn ,

The training set is FFHQ, and the test set is the CelebA-HQ test set. Unfortunately, I don't have the mapping between the indices and the filenames but it should be the normal order of each set. Why do you need the correspondence?

[Kihensarn]

Thanks for your reply!

But I am still confused about the number of training set because there are 70000 samples in the FFHQ while there are 24176 samples in the train_face.pt. Is it a subset of the FFHQ dataset?

And the reason why I want to know the correspondence is that I want to split the dataset you provided according to the attribute annotation file provided by CelebA.

[ssxxx1a]

confused also....

[zhumengzj]

这是来自QQ邮箱的假期自动回复邮件。你好，来信已收到，谢谢！