/U-GAT-IT

Unsupervised Image-to-Image Translation

Primary LanguagePythonMIT LicenseMIT

U-GAT-IT

This an implementation of U-GAT-IT: Unsupervised Generative Attentional Networks with Adaptive Layer-Instance Normalization for Image-to-Image Translation.

Cherry picked examples

examples

Notes

  1. The generator is slightly changed to fit training on a 11 GB graphics card.
  2. Use python train.py for training but set the right configs in model.py first.
  3. Use generate.ipynb for inference after the training.
  4. You can download pretrained (selfie2anime) checkpoints and logs from here.

Credit

This code is based on the official implementation znxlwm/UGATIT-pytorch and taki0112/UGATIT.

Requirements

  1. pytorch 1.3
  2. numpy 1.17
  3. tensorboard 1.15
  4. Pillow 6.1