What dataset was used to train model?

#2
by abhayesian - opened

Was the dataset used private or is it publicly available? Would you be open to sharing the dataset?

Hi, @hadesresearch
I used a private dataset containing approximately 20k images collected from the Internet. Sorry, but I won't share the dataset due to copyright issues.

FYI, there are a couple of publicly available datasets, so you can use them instead.
https://www.gwern.net/Danbooru2021
https://www.kaggle.com/datasets/splcher/animefacedataset

Could you tell me how you cropped the faces? I've seen your landmark detector. Did you just use those boundaries or did you do something else?

@hadesresearch
I applied my face detector to images and used the detected landmarks to align faces to the mean face. You can find the mean face data here. The mean face points are in the normalized coordinates, so you need to shift and rescale them.
I think the code snippet in this issue comment would help you understand how to align faces.

Hope this helps.

Thanks for the help!

abhayesian changed discussion status to closed

Sign up or log in to comment