Turn your videos (and selfies) into anime!
This repository uses a generative adversarial network to transform real-life videos and images into anime. It is based on the paper [1] and the official implementation [2], which contain a lot more stuff. I suppose most people are more interested in trying out the anime generation demo, so I made the following adaptations:
Guy vs. Girl Video Results
Note: training data was biased, containing only images of women, which might explain why things could get weird for guys close-up :laughing:
Checkpoint | Description | Link | Size |
---|---|---|---|
UGATIT_100_epoch_generator_only | Minified, generator only checkpoint based on 100-epoch checkpoint from [2] | Google Drive | 1 GB |
Script | Description | Run (with help for options) |
---|---|---|
record.py | Record your video to anime live with a webcam* | python record.py --help |
selfie.py | Turn your selfie into an anime character | python selfie.py --help |
*Note: powerful computer with NVIDIA GPU may be required
This project is based on the paper [1] and official TensorFlow implementation by the authors [2].
[1] Junho Kim, Minjae Kim, Hyeonwoo Kang, et al. “U-GAT-IT: Unsupervised Generative Attentional Networks with Adaptive Layer-Instance Normalization for Image-to-Image Translation”. In: arXiv preprint arXiv:1907.10830 (2019).