ProgressiveSamplingSchedule
Models & Types · AiDotNet.NeuralRadianceFields.Data
Progressive coarse-to-fine ray-sampling schedule (#1834 excellence goal #4). Reference NeRF
impls hard-code a fixed sample count (64 coarse + 128 fine); nerfstudio requires the schedule
as a YAML entry that only takes effect if the caller wired the appropriate scheduler class
into the training pipeline. Here it’s first-class: instantiate a
ProgressiveSamplingSchedule and hand it to the image-space training call — the
facade selects the coarse count while density is still forming, then ramps to the fine count.
Constructors
| Constructor |
Summary |
ProgressiveSamplingSchedule() |
|
Properties
| Property |
Summary |
int CoarseSamples { get; set; } |
Sample count in the coarse phase (early iterations). |
int FineSamples { get; set; } |
Sample count in the fine phase. |
int RampEndIteration { get; set; } |
Iteration at which the schedule finishes ramping from coarse to fine. |
Methods
| Method |
Summary |
ProgressiveSamplingSchedule Fast() |
Fast schedule — smaller sample counts for prototyping / low-budget training. |
ProgressiveSamplingSchedule Paper() |
Convenience — the paper’s hierarchical schedule: 64 coarse + 128 fine, ramping over 5k iters. |
int SamplesForIteration(int iteration) |
Returns the sample count for iteration iteration. |