ColorTransferLab
- Web-based platform designed for user-friendly application of color and style transfer as well as colorization algorithms across diverse data types.
- Supported data types: images, textured meshes, point clouds, light fields, videos, gaussian splattings, volumetric videos.
- Supports complete self-hosting.
- Support for 11 color transfer, 5 style transfer and 3 colorization algorithms, configurable directly through the web interface.
- Incorporation of multiple objective evaluation metrics for comprehensive analysis.
Abstract
Color manipulation is a significant topic not only in artistic fields, such as film production and photography, but also in research areas such as medical imaging and forensics, where it has been extensively studied. One prominent approach in this domain focuses on automatically transferring color statistics from one image to another, a process known as color transfer. Over the past few decades, a substantial number of algorithms have been developed and refined to perform color transfer on various data types, including images, videos, point clouds, and light fields. To address the limited accessibility and usability of some existing algorithms, we introduce a new version of ColorTransferLab, a web-based software testbed designed for the straightforward application of a wide range of color transfer, style transfer, and colorization algorithms. This platform extends support to textured 3D meshes, volumetric videos, and Gaussian splattings, alongside the previously mentioned data types. It also provides access to objective metrics widely used for evaluating the quality of color transfer results over recent decades, complementing visual assessments. ColorTransferLabV2 is intended to serve as a valuable tool for researchers and artists in the field of automatic color grading, enabling experimentation with color transfer techniques across multiple media.
HueBedroom Dataset
This dataset provides 3D reconstructions of a single room under varying lighting conditions, achieved by illuminating the space with hue color lamps. It serves as a valuable ground truth for evaluating color transfer algorithms.
Sunlit3D Dataset
The Sunlit3D dataset offers post-processed 3D reconstructions of buildings for shadow removal tasks. Existing 3D models with appropriate licenses were sourced from SketchFab, focusing on those with minimal shading. This was crucial, as each model was re-illuminated and the lighting baked into the texture map, providing data suitable for shadow removal. More details are provided in the section below.
Intrinsic Decomposition Dataset
The Interior Decomposition Dataset (IDD) is a dataset designed to analyze how intrinsic scene properties influence the perceived quality of color transfer methods. It consists of photorealistic, synthetically rendered indoor scenes with systematically controlled variations in color, illumination, geometry, object arrangement, and viewpoint. In addition to the rendered images, the dataset provides detailed ground-truth information such as semantic segmentation, depth, and physically-based rendering passes.
Comments