Project Page

Instant Photorealistic Style Transfer (IPST) approach is designed to achieve instant photorealistic style transfer on super-resolution inputs without the need for pre-training on pair-wise datasets or imposing extra constraints. Our method utilizes a lightweight StyleNet to enable style transfer from a style image to a content image while preserving non-color information. To further enhance the style transfer process, we introduce an instance-adaptive optimization to prioritize the photorealism of outputs and accelerate the convergence of the style network, leading to a rapid training completion within seconds. Moreover, IPST is well-suited for multi-frame style transfer tasks, as it retains temporal and multi-view consistency of the multi-frame inputs such as video and Neural Radiance Field (NeRF).

The quickstart will help you install IPST and be familiar with the transfer commands.

An NVIDIA video card with installed CUDA.

git clone https://github.com/RongLiu-Leo/IPST.git
cd IPST

conda create --name IPST -y python=3.8
conda activate IPST
pip install --upgrade pip

pip install -r requirements.txt

For a content-style image pair, run

python transfer.py --content-image {content_image_path} --style-image {style_image_path}

By default, IPST will load and transfer all video frames, which means the batch size is equal to the number of frames. This setting achieves extreme speed but can potentially cause memory problems.

python transfer.py --content-video {content_image_path} --style-image {style_image_path}

Using the --frame-by-frame option will set the batch size to 1, enabling the processing of longer videos at the expense of speed.

python transfer.py --content-video {content_video_path} --style-image {style_image_path} --frame-by-frame True

If it is still not working, try to split the whole video into smaller videos.

@misc{liu2023instant,
    title={Instant Photorealistic Style Transfer: A Lightweight and Adaptive Approach}, 
    author={Rong Liu and Enyu Zhao and Zhiyuan Liu and Andrew Wei-Wen Feng and Scott John Easley},
    year={2023},
    eprint={2309.10011},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}