We present GaSLight, a method that generates spatially-varying lighting from regular images. Our method proposes using HDR Gaussian Splats as light source representation, marking the first time regular images can serve as light sources in a 3D renderer. Our two-stage process first enhances the dynamic range of images plausibly and accurately by leveraging the priors embedded in diffusion models. Next, we employ Gaussian Splats to model 3D lighting, achieving spatially variant lighting. Our approach yields state-of-the-art results on HDR estimations and their applications in illuminating virtual objects and scenes. To facilitate the benchmarking of images as light sources, we introduce a novel dataset of calibrated and unsaturated HDR to evaluate images as light sources. We assess our method using a combination of this novel dataset and an existing dataset from the literature.
HDR reconstruction
Training
Inference
3D lighting representation
Evaluation Datasets
SI-HDR: We share our reconstructions for the clip_95 images of the
SI-HDR dataset
using our gaslight method
here.
We found some HDR images to be saturated
list
BtP-HDR: We adapt the Theta Dataset from
Beyond the Pixel
to obtain a HDR dataset with reference HDR images,
input LDR images directly produced by the camera and
reconstructions from publicly available methods
(ExpandNet,
HDRCNN,
MaskHDR,
SingleHDR)
as well as our own gaslight. The full dataset is available
here.
Citation
@article{bolduc2025GaSLight,
title={GaSLight: Gaussian Splats for Spatially-Varying Lighting in HDR},
author={Bolduc, Christophe and Hold-Geoffroy, Yannick and Shu, Zhixin and Lalonde, Jean-Fran{\c{c}}ois},
journal={ArXiv},
year={2025}
}
Acknowledgements
This research was supported by Sentinel North, NSERC grant RGPIN 2020-04799, and the Digital Research Alliance Canada.
