Color Matching Using Hypernetwork-Based Kolmogorov-Arnold Networks

🇷🇺 Russian language (Русский язык)

Authors: Artem Nikonorov, Georgy Perevozchikov, Andrei Korepanov, Nancy Mehta, Mahmoud Afifi, Egor Ershov, Radu Timofte.

Abstract

We present cmKAN, a versatile framework for color matching. Given an input image with colors from a source color distribution, our method effectively and accurately maps these colors to match a target color distribution in both supervised and unsupervised settings. Our framework leverages the spline capabilities of Kolmogorov-Arnold Networks (KANs) to model the color matching between source and target distributions. Specifically, we developed a hypernetwork that generates spatially varying weight maps to control the nonlinear splines of a KAN, enabling accurate color matching. As part of this work, we introduce a large-scale dataset of paired images captured by two distinct cameras to evaluate our method’s efficacy in matching colors produced by different cameras. We evaluated our approach across various color-matching tasks, including: (1) raw-to-raw mapping, where the source color distribution is in one camera’s raw color space and the target in another camera’s raw space; (2) raw-to-sRGB mapping, where the source color distribution is in a camera’s raw space and the target is in the display sRGB space, emulating the color rendering of a camera ISP; and (3) sRGB-to-sRGB mapping, where the goal is to transfer colors from a source sRGB space (e.g., produced by a source camera ISP) to a target sRGB space (e.g., from a different camera ISP). The results demonstrate that our method achieves state-of-the-art performance across these tasks while remaining lightweight compared to other color matching and transfer methods.

Installation and Requirements

Create a conda (or python) environment, clone the repository and install the required packages:

# 1. Create an environment

conda create -n cmKAN python=3.13 pip
conda activate cmKAN
# or
python -m venv .venv
source .venv/bin/activate

# 2. Clone the repository
git clone https://github.com/gosha20777/cmKAN.git
cd cmKAN

# 3. Install the required packages
pip install -r requirements.txt

Dataset

We introduce a large-scale Volga2K dataset, featuring over 2000 well-aligned images captured with a Huawei P40 Pro. This device was selected for its distinct Quad-Bayer RGGB and RYYB camera sensors, which employ different image processing techniques. These sensor variations in color handling, sensitivity, and tone-mapping create a significant domain gap. Spanning four years and multiple locations, the dataset is designed to effectively evaluate color-matching methods.

You can find more details about the Volga2K dataset on 😡 Hugging Face 😡 official page. To download Volga2K, please use the following command:

huggingface-cli download gosha20777/volga2k --repo-type dataset --local-dir data/

Pre-trained Models

We provide pre-trained models for the following tasks:

Dataset	Task	Training Config	Checkpoint
Volga2K	sRGB-to-sRGB	unsupervised	checkpoint
Volga2K	sRGB-to-sRGB	supervised	checkpoint
Volga2K	sRGB-to-sRGB	pair-based	checkpoint
Adobe 5K (password: 5fyk)	sRGB-to-sRGB	supervised	checkpoint
Samsung2Iphone	raw-to-raw	unsupervised	checkpoint
Zurich raw-to-sRGB	raw-to-sRGB	supervised	checkpoint

How To Use

Our cmKAN provides a command-line interface (CLI) to interact with the following tools:

python main.py -h 

Usage: cmKAN CLI [-h] {data-create,test,train,predict,unit-test} ...

Options:
  -h, --help            Show this help message and exit

Tools:
  {data-create,test,train,predict,unit-test}
    
    data-create         Create dataset
    train               Train model
    test                Test model
    predict             Run model inference
    unit-test           Run unit tests

For all the tools, you can use the -h flag to get help on how to use them (e.g. python main.py train -h). Here are some examples on how to use the tools:

Train

python main.py train -c configs/config.yaml

Test

python main.py test -c configs/config.yaml -w checkpoint.ckpt

Predict

python main.py predict -c configs/config.yaml -i path/to/input/folder -o path/to/output/folder

Additional Guides

You can find additional guides how to reproduce all our experiments in our wiki page. We provide detailed instructions on how to train and test our model, as well as how to use it for inference.

Citation

If you find our work useful, please consider citing it:

@article{perevozchikov2025color,
  title={Color Matching Using Hypernetwork-Based Kolmogorov-Arnold Networks},
  author={Nikonorov, Artem and Perevozchikov, Georgy and Korepanov, Andrei and Mehta, Nancy and Afifi, Mahmoud and Ershov, Egor and Timofte, Radu},
  journal={arXiv preprint arXiv:2503.11781},
  year={2025}
}