UAVLight: A Benchmark for Illumination-Robust 3D Reconstruction in Unmanned Aerial Vehicle (UAV) Scenes

TL;DR

UAVLight addresses the challenge of illumination variability in outdoor UAV-based 3D reconstruction by providing a controlled-yet-real benchmark with repeatable waypointed trajectories captured at multiple times of day. It combines high-quality, geo-referenced geometry with standardized splits and sun-ground-truth annotations to enable fair evaluation of reconstruction methods under time-varying lighting. The framework supports both implicit appearance models and explicit inverse-rendering approaches, revealing that explicit lighting decomposition generally yields better disentanglement and relighting stability across illumination slots. This benchmark advances the field by offering a realistic yet controllable testbed for illumination-robust reconstruction and relighting, with potential for future expansion to richer atmospheric conditions and outdoor simulations.

Abstract

Illumination inconsistency is a fundamental challenge in multi-view 3D reconstruction. Variations in sunlight direction, cloud cover, and shadows break the constant-lighting assumption underlying both classical multi-view stereo (MVS) and structure from motion (SfM) pipelines and recent neural rendering methods, leading to geometry drift, color inconsistency, and shadow imprinting. This issue is especially critical in UAV-based reconstruction, where long flight durations and outdoor environments make lighting changes unavoidable. However, existing datasets either restrict capture to short time windows, thus lacking meaningful illumination diversity, or span months and seasons, where geometric and semantic changes confound the isolated study of lighting robustness. We introduce UAVLight, a controlled-yet-real benchmark for illumination-robust 3D reconstruction. Each scene is captured along repeatable, geo-referenced flight paths at multiple fixed times of day, producing natural lighting variation under consistent geometry, calibration, and viewpoints. With standardized evaluation protocols across lighting conditions, UAVLight provides a reliable foundation for developing and benchmarking reconstruction methods that are consistent, faithful, and relightable in real outdoor environments.

Figures (6)

• Figure 1: Overview of the UAVLight benchmark. Each scene is captured by low-altitude UAV flights along fixed waypointed trajectories at multiple times of day. Our benchmark records natural illumination changes along consistent geometry and viewpoints, enabling quantitative evaluation of illumination-robust reconstruction and relighting.
• Figure 2: Visualization of representative dense point clouds from 12 selected scenes in our benchmark.
• Figure 3: Illustration of illumination variations across similar viewpoints at different times of day. The bottom-right panel shows the corresponding ground-truth sunlight directions computed from GPS and timestamps.
• Figure 4: Paired cross-light protocol used for evaluating methods. For each time slot, lighting is estimated from one subset of views and applied to another subset captured under the same time slot.
• Figure 5: Visualization of the reconstruction results from different baselines on five UAVLight scenes.