IDensificationSchedule
Interfaces · AiDotNet.NeuralRadianceFields.Data
Densification firing schedule for GaussianSplatting (#1835 excellence goal #2). Reference GS impls hard-code a fixed-interval schedule (“every 100 iters densify”); AiDotNet exposes this as a swappable strategy so callers can plug in a data-driven schedule (gradient-variance / loss-plateau triggers) without touching the model.
Known implementations: AdaptiveDensificationSchedule, FixedIntervalDensificationSchedule
Methods
| Method | Summary |
|---|---|
bool ShouldFire(int iteration, double lastLoss, double lastGradientNorm) |
Whether densification should fire at iteration given the observed signals. |