VFHQ dataset download issue

[jseobyun]

Hello the authors. I appreciate for sharing the data.

As following the process, I have some problems when downloading VFHQ dataset.

1) VFHQ Download script fails.

In provided "download_vfhq.py" file, vid_path = ydl_opts["outtmpl"] at line 167 seems weird.

It returns dictionary filled with many keys and values, so following ffmpeg cli does not work correctly with not found error.

I think it should be modified to indicate newly saved video path from ydl.

I temporally revise it to work correctly but official support will be more helpful :)

Could you check downlaod_vfhq.py file? If it is ydl version issues, please let us know the version.

2) VFHQ annotations are different with id_train.txt and id_test.txt

I found the annotations you explained in official VFHQ repository, meta_info. However, it only provides 15381 items.

In id_train.txt and id_test.txt, there are more than 16000 items. Where can I find the annotations that used in EFHQ?

3) Storage issue

Actually It is not an issue but the storage required for following the process is huge. I think more than 10 TB should be prepared for covering all videos and images at the same time.

Is it essential saving all videos at disk? How about deleting each video after extracting images?

Thanks in advance,
Best regards.