Volga2K dataset

In our cmKAN paper, we presented a large-scale Volga2K dataset captured using a Huawei P40 Pro phone wich containts 1263 well well-aligned image paires (more than 2K images in total). This device was specifically chosen because it features two distinct cameras with different sensor types: Quad-Bayer RGGB sensor (Sony IMX700) and RYYB sensor (Sony IMX608). These differences in sensor types result in varying image processing algorithms, with the RGGB and RYYB sensors requiring distinct demosaic methods, white balance (WB) corrections, and color calibration. The RYYB sensor, being more sensitive to light and affected by color lens shading (caused by small focal length and wide lens angle), also demands different tone-mapping techniques. Furthermore, the color gamut and saturation levels differ significantly between RGB and RYB, especially under low-light conditions. These variations in sensor behavior introduce a substantial domain gap between images captured by the two cameras, which is ideal for evaluating color-matching methods. The dataset thus simulates the real-world challenge of color matching between cameras.

Our dataset, spanning four years of data collection, includes images captured at over 20 locations across four countries. The images are well-aligned with minimal matching errors, and we provide detailed annotations for various scenes. Additionally, our dataset includes keypoint clouds generated using the SURF detector for each image pair, along with binary matching masks to exclude misaligned regions. While our proposed method performs well even with unaligned data, this additional information is valuable for future research. To ensure ease of access and usability, we also offer a command-line interface (CLI) toolkit for processing, preparing, and cropping the dataset into smaller sections, making it more convenient for use in a variety of tasks.

Dataset statistics

Scene class	Count	Image class	Count
Indoor	114	City	324
Spring	109	Countryside	164
Summer	558	Forest	49
Autumn	175	Mountains	166
Winter	97	Seaside	161
Low-light	107	Sunset	92
Total 1260	Low-light/Difficult illumination	21

Dataset structure

The dataset contains the following file types:

• source images, from the Huawei P40 Pro's Sony IMX700 sensor
• aligned target (or reference) images, from the Huawei P40 Pro's Sony IMX608 sensor
• matched color components from Sony IMX700 and Sony IMX608 sensors, saved as 6*N numpy arrays, where first three columns represent the source RGB colors and last three columns represent the target RGB colors. The color matching is based on FastSLIC superpixel segmentation, with additional filtering to remove non-uniform superpixels or superpixels exhibiting very strong color differences. Remark: In our cmKAN approach we do not use these numpy color components, but they can be useful for evaluating classical color-matching methods (e.g. linear regression, color-homography, etc.).

The dataset is organized into three main folders: source, reference, and feature. The source folder contains the source images, the reference folder contains the aligned reference images, and the feature folder contains the keypoint clouds. The dataset is structured as follows:

volga2k/
├── feature    # Matched color components
│   ├── P40_S0278_src_m.npy
│   ├── P40_S0281_src_m.npy
│   └── ...
├── reference  # Aligned reference (target) images
│   ├── P40_S0278_ref_m.jpg
│   ├── P40_S0281_ref_m.jpg
│   └── ...
└── source     # Source images
    ├── P40_S0278_src_m.jpg
    ├── P40_S0281_src_m.jpg
    └── ...

Data processing

In our cmKAN repository, we provide a command-line interface (CLI) toolkit for processing, preparing, cropping and splitting the dataset into training, validation and testing sets. The toolkit includes the following commands:

python main.py data-create -h