New scientific paper using Pepper&Carrot sources 👍 "Deep Normal Estimation for Automatic Shading of Hand-Drawn Characters by Trinity College Dublin"

@davidrevoy The end result looks as if it's just an embossed image though. May be, with a better (deeper) normal map it will do better, but I think the point is the normal map algorithm can't tell the difference between lines that define volumes and shapes and lines that define detail (and thus don't need to be normal-mapped, or at least done differently).

Still impressive though.

@drequivalent Exactly. It still looks like pillow with sewing; or ballons. But I had a small discussion this morning with the lead author and they get already better result with more training of the neural network. (eg: to detect a face; and keep the eye flat while shading as a whole the head sphere). For sure, this type of studies are just the beginning of machine learning how to correctly shade 2D artworks. I can't imagine how this will evolve in 10 or 20 years :D

Inscrivez-vous pour prendre part à la conversation

Le réseau social de l'avenir : Pas d'annonces, pas de surveillance institutionnelle, conception éthique et décentralisation ! Possédez vos données avec Mastodon !