Relighting from a Single Image. Datasets and Deep Intrinsic-based Architecture

Yixiong Yang, Hassan A. Sial, Ramon Baldrich, and Maria Vanrell
Computer Vision Center, Universitat Autònoma de Barcelona
IEEE Transactions on Multimedia (TMM), 2025
Paper Code

ISR: Intrinsic Scene Relighting Dataset

The images of this dataset are generated by the open-source Blender rendering engine. This dataset includes intrinsic components in the ground truth. The ISR has 7,801 scenes, each one under 10 different light conditions. Each scene has between 3 and 10 non-overlapping objects which are randomly selected from various categories of the ShapeNet dataset including electronics, pots, buses, cars, chairs, sofas, and airplanes. For more details, please check our paper and GitHub repo.

Illustrations of some examples from the dataset, showing the images along with their corresponding reflectances and shading. We also show the coordinate system employed in the dataset, which includes the pan and tilt.

Download ISR

The dataset can be downloaded separately by component:

You can download only the Reflectance and Shading, as the image can be calculated from them in the code.

Note: The filename format is: {scene_part1}_{scene_part2}_{pan}_{tilt}_{light_temperature}_{unused}. The first two numbers represent the scene ID, while pan, tilt, and light temperature represent the lighting parameters.

RSR: Real Scene Relighting Dataset

To validate the effectiveness of our method in real-world scenes, we developed the real scene-relighting (RSR) dataset. The RSR dataset is acquired in a lab with an automatic setup. We use a DSLR Nikon D5200 camera to capture the images. The scene area is lighted with a 3 × 3 multicolored LED light matrix. The scene is set on a rotatable table. Each scene is composed of a background and several objects. The library of backgrounds has 8 textured options. We carefully choose a set of 80 real-life objects sampling a wide range of materials and shapes. From glossy to matte materials and from natural to basic shapes. The objects are organized into 3 groups: 20 natural (organic and stone), 40 manufactured (plastic, paper, glass, or metal), and 20 basic (foam) shapes.

We show the platform for building the dataset, the collection of objects in the dataset, and some examples with different light positions and colors.

Download RSR

Note: The filename format is: {picture_idx}_{group_idx}_{pan}_{tilt}_{R}_{G}_{B}_{scene_idx}_{unused}_{view_idx}_{light_position_idx}. The quantities that need attention are pan, tilt, and color (RGB), which represent the parameters of the light. The order of the lights is as follows:

5 4 3
6 1 2
7 8 9

Citation

If you find this work useful for your research, please cite:

@article{yang2025relighting,
  title={Relighting from a Single Image: Datasets and Deep Intrinsic-based Architecture},
  author={Yang, Yixiong and Sial, Hassan A. and Baldrich, Ramon and Vanrell, Maria},
  journal={IEEE Transactions on Multimedia},
  year={2025},
  doi={10.1109/TMM.2025.3537888}
}

Acknowledgments

This work was supported by grants PID2021-128178OB-I00 funded by MCIN/AEI/10.13039/501100011033 and by ERDF "A way of making Europe", and by the Generalitat de Catalunya CERCA Program.