onion-liu/aahq-dataset

Artstation-Artistic-face-HQ Dataset (AAHQ)

Hi there 👋! Here is onion 🧅!

aahq-dataset's Issues

download error

requests.exceptions.ConnectionError: HTTPSConnectionPool(host='cdnb.artstation.com', port=443)
I try some solutions like 'https://blog.csdn.net/weixin_42210687/article/details/103480208', but still not work well. Sorry to bother you that do you know this prob?
thanks.

Several files not found

Hi, Thanks for sharing dataset

I found several (=121) files are not exist while downloading the dataset.
Could you share currently not available images?

I add some not founded urls. You can open it using pickle.

notfound.txt

How to get URLs from Artstation

Thanks for your great work! I think this dataset contributes a lot to the machine learning community and is also useful to me. I am wondering how you collected URLs from a specific channel (e.g., portraits) in Artstation. Can you share your code?

One image missing and it interrupted the download

I download the dataset several times. It always terminal on the following image:

FileNotFoundError: [Errno 2] No such file or directory: './raw\033_811_848_4k_elina-mei-rz75hvsungrap8uuohcswbpuqlidf3rxmq7ehpycvqpb5pwzyreu2s3hc7tsqjfns9pekzogzfcevvdzdcrwqyb2kaqidmbzmksxyfnhw1ddfxqa6gmvowbgakw9soj3t4dugditsv7sunzzxakqysinxjd0yxg1xe7bebqatuyvt0mcgl21ejhnwdplxnwtb5oe3uwf.jpg

Question: why first transform() to large size then resize() to smaller size?

With the default arguments in image_align(), you first transform to a large size (4096x4096) and then resize to a smaller size (1024x1024). Why not just directly transform into the smaller size?

BTW, thank you so much for this awesome tool!

Split into training and evaluation sets

Thanks for your great work! I got a question about this dataset.

When training BlendGAN, did you split AAHQ dataset into training and evaluation sets and use the training set to train?