NeRFs and Inverse Rendering
Stub. Neural radiance fields (NeRFs): representing a 3D scene as a learned radiance field, the rendering function that turns it into images (and the inverse problem of learning the field from images), and non-visual extensions such as wireless NeRF.
A NeRF represents a 3D scene implicitly, as a neural network mapping (position, viewing direction) to (color, density), and a differentiable rendering function integrates this field along camera rays to produce an image. Fitting a NeRF to a set of posed photographs is itself an inverse problem — recovering the field from its rendered projections — which is what connects it to the rest of the course’s inverse-problem machinery. Wireless NeRF extends the same idea beyond visible light, learning a field over radio signal propagation instead of radiance.
Outline (TODO — flesh out each)
- Radiance fields — scene representation
- The rendering function; NeRF fitting as an inverse problem
- Wireless NeRF