Re-implement CycleGAN in Tensorlayer
Prerequisites:
- Tensorlayer
- TensorFlow
- Python
Run:
CUDA_VISIBLE_DEVICES=0 python main.py
(if datasets are collected by yourself, you can use dataset_clean.py or dataset_crop.py to pre-process images)
Theory:
The generator process:
The discriminator process:
Result Improvement
- Data augmentation
- Resize convolution[4]
- Instance normalization[5]
data augmentation:
Нормализация экземпляра (сравнение с исходной статьейАР Вест V.org/ABS/1607.08…:
Resize convolution (Remove Checkerboard Artifacts):
Final Results:
Reference:
- [1] Original Paper: АР Вест V.org/PDF/1703.10…
- [2] Original implement in Torch: GitHub.com/military training/C-сторона…
- [3] TensorLayer by HaoDong: GitHub.com/это донхао/он…
- [4] Resize Convolution: distill.universal/2016/deco-женский…
- [5] Instance Normalization: АР Вест V.org/ABS/1607.08…