A developer trained an AI model to convert people's photos into cartoons in the style of South Korean cartoonist Lee Mal-nyeon. A GitHub user by the name of bryandlee created a dataset of cartoon faces and trained a face generating model in just ten hours using 500 annotated images.
- The developer incorporated FreezeD, the baseline for transfer learning of GANs, and U-GAT-IT, an image-to-image translation method.
- The code for FreezeG is on GitHub.
- Details on the training can be found on the malnyun_faces project page.
- A Reddit user said it was the first time they had seen when a network "actually understands what it is seeing and redraws it in a different style."