Physically Controllable Relighting of Photographs
Web page · https://yaksoy.github.io/PhysicalRelighting
First seen by Oriel on .
AI summary
This SIGGRAPH 2025 work presents a self-supervised relighting pipeline that gives photographs explicit, physically based light control while preserving photorealistic results through neural rendering.
- The system infers a colored mesh from monocular geometry and intrinsic components, allowing users to define illumination configurations directly in 3D.
- Users can add spot lights, point lights, or environmental illumination, then render the relit scene with a path-tracing engine.
- A feed-forward neural renderer transforms the approximate physically rendered image into the final photorealistic relighting result.
- Differentiable rendering reconstructs scene illumination from real-world photographs, enabling self-supervised training on raw image collections for in-the-wild generalization.
- The approach aims to provide Blender-like explicit lighting control for relighting photographs without requiring traditional 3D scene creation.
#image relighting #neural rendering #differentiable rendering #computer vision
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Hacker News: 66, latest observation ; +59 over observed period since
Comments
Comment count reported by the source.
- Hacker News: 15, latest observation ; +15 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hacker News | 50452224 | https://yaksoy.github.io/PhysicalRelighting/ Read discussion |