SceneBoundsEstimator
Helpers & Utilities · AiDotNet.NeuralRadianceFields.Data
Estimates a tight scene bounding box + near/far ray sampling bounds from a set of camera poses (#1834 excellence goal #3). Reference NeRF pipelines require users to supply near/far and scene bounds by hand — nerfstudio’s config YAML has explicit fields, tiny-cuda-nn requires them at construction. Here it’s automatic: the ImageView pose set is the source of truth, we intersect view frusta to bound the scene tightly, and derive per-view near/far that hit the bbox comfortably from every camera.
Methods
| Method | Summary |
|---|---|
SceneBounds EstimateFromViews(IEnumerable<ImageView<T>> views, double margin) |
Computes the axis-aligned bounding box that contains every view’s principal ray intersection region plus a symmetric buffer around the pose centroid — a paper-quality default when no explicit bounds are supplied. |