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
Notes
- The generator is slightly changed to fit training on a 11 GB graphics card.
- Use
python train.py
for training but set the right configs inmodel.py
first. - Use
generate.ipynb
for inference after the training. - 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
- pytorch 1.3
- numpy 1.17
- tensorboard 1.15
- Pillow 6.1