SkyLume is a large-scale real-world UAV dataset for urban scene reconstruction under varying illumination. It captures the same urban regions across morning, noon, and afternoon flights, pairing high-resolution five-direction UAV imagery with LiDAR-derived geometry for robust 3D reconstruction, novel view synthesis, inverse rendering, and cross-time consistency research.
Project page:
https://zhuoxiaoli.github.io/skylume_page/
Paper: Beyond a Single Light: A Large-Scale… See the full description on the dataset page:
https://huggingface.co/datasets/vavaw/SkyLume.
