I am Vijish Madhavan, My experience is into marketing. Coding ka kida caught me last year.
This is just a fun project!!
Introducing Artline, A Deep Learning based project for creating line art portraits.
As I said it was just a fun project, but the response I got surprised me. The results seems to be state of the art comparing it to previous models like APDrawingGan and U-2-Net. You are free to judge as its opensource.
GitHub Link
Check out for pretty images, scroll through the repo link.
https://github.com/vijishmadhavan/ArtLine
Gist of the project
Note: Being a newbie my code and documentation would be bad, I will improve with time.
Technical Details
Self-Attention (https://arxiv.org/abs/1805.08318). Generator is pretrained UNET with spectral normalization and self-attention. Something that I got from Jason Antic's DeOldify(https://github.com/jantic/DeOldify), this made a huge difference, all of a sudden I started getting proper details around the facial features.
Progressive Resizing (https://arxiv.org/abs/1710.10196),(https://arxiv.org/pdf/1707.02921.pdf). Progressive resizing takes this idea of gradually increasing the image size, In this project the image size were gradually increased and learning rates were adjusted. Thanks to fast.ai for intrdoucing me to Progressive resizing, this helps the model to generalise better as it sees many more different images.
Generator Loss : Perceptual Loss/Feature Loss based on VGG16. (https://arxiv.org/pdf/1603.08155.pdf).
Surprise!! No critic,No GAN. GAN did not make much of a difference so I was happy with No GAN.
You can follow me on twitter @Vijish68859437
Top comments (0)