What are Radiance Fields?
Radiance fields describe a wide range of photogrammetric reconstruction techniques including Neural Radiance Fields (NeRFs), and 3D-Gaussian splatting (3DGS), that aim to accurately recreate digital versions of real-world objects and environments. Radiance fields use a volumetric approach to reconstruction where millions of discontinuous points are distributed throughout a 3D volume and then an iterative process compares the points to real-world photos or video frames to add, remove, or modify the shape, colour, and distribution of points till they accurately represent the original subject.
Radiance fields differ from the continuous geometric reconstructions that power other tools like the Virtual Soils Monolith Collection (monoliths.soilweb.ca/). Radiance Fields allow for small features (e.g. leaves and roots) to exist in the same model as large features (e.g. trees and mountains) as well as accurate reconstructions of view dependent properties such as the moving reflection of the sun on still water.
The approach to radiance fields that Virtual Soils uses are 3DGS. This was chosen for the wide support provided by free and open-source tools like Lichtfeld Studio (lichtfeld.io/) for model generation and SuperSplat (superspl.at/) for model editing that allowed for simple production of high-quality models that can be viewed in real-time on a wide range of devices from mobile phones to high end desktop PCs.
