Color Transfer with Modulated Flows (AAAI 2025)

arXiv: https://arxiv.org/abs/2503.19062

This is the official implementation of AAAI 2025 paper "Color Transfer with Modulated Flows".

The paper was also presented at "Workshop SPIGM @ ICML 2024".

Please refer to the

• ModFlows_demo.ipynb to use the pretrained model for color transfer on your own images with the demo jupyter notebook
• ModFlows_demo_batched.ipynb to use the pretrained model for color transfer for large images
• HuggingFace for the model checkpoints
• src directory for models definitions
• generate_flows_v2 script for training the dataset of rectified flows
• train_encoder_v2 script for training the encoder

How to clone and download pre-trained weights:

git clone https://github.com/maria-larchenko/modflows.git
cd modflows; git clone https://huggingface.co/MariaLarchenko/modflows_color_encoder

Call python3 run_inference.py --help to see a full list of arguments for inference. Ctrl+C cancels the execution.

Citation

If you use this code in your research, please cite our work:

@article{Larchenko_Lobashev_Guskov_Palyulin_2025, title={Color Transfer with Modulated Flows}, volume={39}, url={https://ojs.aaai.org/index.php/AAAI/article/view/32470},
DOI={10.1609/aaai.v39i4.32470},  number={4},
journal={Proceedings of the AAAI Conference on Artificial Intelligence},
author={Larchenko, Maria and Lobashev, Alexander and Guskov, Dmitry and Palyulin, Vladimir Vladimirovich}, year={2025}, month={Apr.}, pages={4464-4472} }