← AI Terminology
NeRF - Neural Radiance Fields
NeRF represents a 3D scene as a neural network mapping 3D coordinates + view direction to colour and density, rendered via volumetric ray marching.
It sparked the neural 3D scene representation revolution.
It sparked the neural 3D scene representation revolution.
Why It Matters in AI
Photoreal novel-view synthesis from images transformed graphics and vision research. NeRF led to instant variants, mobile capture, and the path toward Gaussian splatting alternatives.
Key Points
| Aspect | Description |
|---|---|
| Map | MLP: (x,y,z,dir) → RGB + σ |
| Use | VFX, heritage, robotics simulation assets |
| Limits | Slow classic training/render; addressed by later methods |
| Origin | Mildenhall et al., 2020 |
| Render | Integrate along camera rays |
| Related | Gaussian splatting, 3D reconstruction, view synthesis |
Simple Analogy
Storing a sculpture not as polygons but as a recipe that tells you the colour and opacity of every point in space, then photographing it from new angles.
Common Usage Examples
- Original NeRF paper demos
- Instant-NGP faster training
- Multi-view capture pipelines
- Compare quality/speed to 3DGS
Summary
In short: NeRF encodes 3D scenes in a neural field for photoreal novel views — the landmark neural scene representation method.