IGARSS 2026 Student Competition

Reconstructing Across Illumination
A Synthetic Dataset for UAV 3D Reconstruction under Varying Light Conditions

IllumUAV-Sim provides 19,836 high-resolution images with pixel-aligned day-night pairs and complete geometric ground truth for advancing nighttime UAV perception research.

19,836
Total Images
45MP
Resolution
6
Lighting Conditions
2 km²
Scene Coverage

Research Overview

IllumUAV-Sim is a high-fidelity synthetic dataset designed for UAV 3D reconstruction under diverse illumination conditions. Built with Unreal Engine 5, the dataset provides 19,836 high-resolution images (8196×5460 pixels) with pixel-aligned views captured under six distinct lighting conditions that span the full day–night cycle. Each frame is annotated with dense geometric ground truth, including depth maps, surface normals, and camera parameters. By offering comprehensive coverage of illumination variation and accurate geometric annotations, IllumUAV-Sim enables robust evaluation and training of methods for low-light UAV perception and 3D reconstruction.

IllumUAV-Sim Dataset

The first synthetic dataset designed specifically for UAV 3D reconstruction under Varying Light Conditions

Pixel-Aligned Pairs

Perfect spatial correspondence between day and night imagery from identical camera poses, enabling supervised training.

Complete Annotations

Per-pixel depth maps, surface normals, camera parameters, and illumination metadata for each frame.

Diverse Urban Scene

2 km² photorealistic city environment with residential, commercial buildings, roads, and Highway network.

High Resolution

8196×5460 pixels (~45MP) per image, supporting high-fidelity reconstruction research.

UAV Configurations

Nadir (551) and oblique (2,755) imaging positions with 80%/60% forward/side overlap.

Full Day-Night Cycle

Six illumination conditions from noon to late night covering complete diurnal variations.

Illumination Conditions

Same viewpoint rendered under six different lighting configurations

Noon Illumination
Noon
Afternoon Illumination
Afternoon
Dusk Illumination
Dusk
Early Night Illumination
Early Night
Late Night Illumination
Late Night
Base Color
Base Color

Get the Data

Access sample data, code, and generation tutorials

Code & Tools

Evaluation scripts and data path planning tools

GitHub
Code

Sample Data

Representative samples from each illumination

Download
~ 1.6 GB

Generation Tutorial

Tutorial to generate data

Tutorial
Documentation