Get Started
Home
Topics
Search
Library
Research questionHow can we relight driving images continuously across day and night while preserving scene structure without paired data?Driving datasets often capture scenes under limited lighting conditions, while relighting must change illumination without altering the scene’s identity or structure. The lack of paired views of the same scene under different lighting makes continuous day–night translation difficult to learn.
AI
Computer Vision
Diffusion Models
Image & Video Processing
Image Generation
Machine Learning
Latest papersRecent research connected to this question, newest first.HELIOS: From midnight to noon, continuous outdoor urban scene relightingThe source addresses outdoor urban driving images and night-to-day and day-to-night relighting trained on unlabeled real-world datasets without paired images. It describes GPS-derived solar-angle control, albedo conditioning and distillation, and a cycle-consistent diffusion pipeline; evidence comes from the reported evaluations and user study.research paper · Sep 2, 2026
Related questions
How can self-supervised monocular depth estimation stay accurate across scales and dynamic traffic on automotive edge devices?How can in-camera colorimetric mapping stay accurate under non-Planckian LED illumination?How can end-to-end autonomous-driving world models use surround-view inputs without costly future-video generation at inference?How can HD map construction remain reliable with missing camera views and inaccurate camera poses?