Self-Supervised Pipeline Enables Physically Controllable Photo Relighting

1 reports archived1 independent sourcesupdated 11/11/2026, 19:38:43

Synthesis & Latest Updates

1 Sources Cross-Validated

The paper introduces a self-supervised approach to in-the-wild image relighting that combines the physical accuracy of traditional rendering with the photorealism of neural rendering. A colored mesh inferred from monocular geometry and intrinsic components lets users define 3D lighting, which is path-traced and then refined by a feed-forward neural renderer trained via a differentiable rendering process.

LATEST/The paper introduces a self-supervised approach to in-the-wild image relighting that combines the physical accuracy of traditional rendering with the photorealism of neural rendering. A colored mesh inferred from monocular geometry and intrinsic components lets users define 3D lighting, which is path-traced and then refined by a feed-forward neural renderer trained via a differentiable rendering process.

Hourly heat curve

7 fully observed hours

Now

12.1

Peak

13.922:00

24h change

–

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 1 reports · Latest first

Hacker News AIT2Nov 11, 19:38·72 pts
Self-Supervised Pipeline Enables Physically Controllable Photo Relighting
- Users explicitly place and color lights in 3D rather than relying on text or reference-image prompts
- Differentiable rendering recovers in-the-wild illumination, enabling self-supervised training on raw image collections
- Goal is to give ordinary photographs the light-editing control of 3D graphics tools
View Related Reports & Analysis Original Source