Joint Bilateral Learning

This repository is an unofficial implementation in PyTorch for the paper: "Joint Bilateral Learning for Real-time Universal Photorealistic Style Transfer" (ECCV2020)

[paper]

Dependencies

• Python 3.7.2
• PyTorch 1.2
• CUDA10.0 and cuDNN
• pytorch-lightning(newest, 1.04rc1)

Train

$ python main.py --cont_img_path <path/to/cont_img> --style_img_path <path/to/style_img> --batch_size 8

Test

$ python test.py --cont_img_path <path/to/single_cont_img> --style_img_path <path/to/single_style_img> --model_checkpoint <path/to/model checkpoint>

Use pytorch-lightning

• Train

python main_pl.py --gpus=2,3,4,5 --batch_size=6 --num_workers=6 --vgg_ckp_file=<vgg checkpoint file> --content_path=<content images path> --style_path=<style images path> --val_content=<validation content images path> --val_style=<validation style images path>