Interface IGaussianSplattingStreamOptions
debugDisplay?: boolean;
debugLodSource?: GaussianSplattingStreamDebugLodSource;
decodeSh?: boolean;
deflateURL?: string;
evictionCooldownFrames?: number;
fflate?: any;
frustumCulling?: boolean;
lodBaseDistance?: number;
lodBehindPenalty?: number;
lodCooldownFrames?: number;
lodMultiplier?: number;
lodRangeMax?: number;
lodRangeMin?: number;
lodUpdateDistance?: number;
lodUpdateInterval?: number;
maxConcurrentDownloads?: number;
maxDecodesPerFrame?: number;
maxDetailLod?: number;
maxDownloadRetries?: number;
maxResidentSplats?: number;
memoryBudgetMb?: number;
needsRotationScale?: boolean;
splatBudget?: number | "auto";
}
Index
Properties
Properties
Optionaldebug
When true, renders a wireframe box per LOD node, colored by the node's LOD level.
Optionaldebug
Which LOD value drives the debug wireframe colors. Defaults to "optimal".
Optionaldecode
When true, higher-order spherical-harmonics carried by the SOG files (shN) are GPU-decoded into baked
packed-u32 SH textures so the streamed splats render with view-dependent lighting (matching the non-stream
.spz/.sog path) instead of flat DC-only color. Streams with complete initial metadata use the maximum
source SH degree. Streams that defer finer metadata reserve the supported degree-4 layout before it is known,
even if the files ultimately carry no shN; lower-degree files neutral-fill the unused bands.
Defaults to true; set to false to force flat DC-only color and avoid the SH decode cost/texture memory.
Optionaldeflate
URL of the fflate UMD module used to unzip .sog environment bundles.
Optionaleviction
Frames an unreferenced (no longer rendered) LOD file stays resident before it is evicted, so a quick
return to it avoids a re-download. Only used when a budget enables eviction. PlayCanvas default 100.
Optionalfflate
Pre-loaded fflate module.
Optionalfrustum
When true (default), LOD nodes outside the camera frustum are biased to their coarsest LOD rather than
rendered at full detail. They stay in the sort/render set so they appear instantly (at low detail) when
the camera turns toward them, then refine. Set to false to render every node at its distance LOD.
Optionallod
Distance (in local units) of the first LOD transition. PlayCanvas default 5.
Optionallod
Distance multiplier applied to nodes behind the camera (1 = no penalty). PlayCanvas default 1.
Optionallod
Frames a node must wait after switching LOD before it may switch again (oscillation damping). Defaults to 10.
Optionallod
Geometric ratio between successive LOD transition distances. PlayCanvas default 3.
Optionallod
Highest LOD index the optimal-LOD heuristic may select. Defaults to lodLevels - 1.
Optionallod
Lowest LOD index the optimal-LOD heuristic may select. Defaults to 0.
Optionallod
Minimum camera movement (world units) required to re-evaluate LODs. Defaults to 0.5.
Optionallod
Minimum number of frames between LOD re-evaluations (throttles per-frame work during motion). Defaults to 4.
Optionalmax
Maximum number of LOD file downloads allowed to run concurrently. PlayCanvas default 2.
Optionalmax
Maximum number of LOD source files to GPU-decode per frame (spreads work to avoid hitches). Defaults to 1.
Optionalmax
Finest (most detailed) LOD level any node is allowed to render. 0 allows full detail (level 0);
1 caps detail at the next-coarser level, and so on. Higher values force a coarser maximum detail.
Optionalmax
Number of times a failed file download is retried before giving up. PlayCanvas default 2.
Optionalmax
Initial maximum number of splats kept resident in the fixed-size work buffer. It is raised when necessary to fit the complete coarse layer. When unset and the complete source size is known, streams retain the complete source. Streams that defer finer metadata use a bounded device-tiered default with coarse-derived refinement headroom. Finite non-positive values leave this limit unset; non-finite or unsafe positive counts are rejected. Positive fractional counts are floored to at least one splat before applying the coarse minimum.
Optionalmemory
Initial GPU/CPU memory estimate (in megabytes) for resident splats. It is converted to a splat count using the per-splat cost (core data plus any baked SH and rotation/scale textures), combined with maxResidentSplats by taking the smaller limit, then raised when necessary to fit the complete coarse layer. The work buffer has a fixed lifetime capacity; construct a new stream to use another limit. Finite non-positive values leave this limit unset; non-finite values are rejected.
Optionalneeds
When true, each splat's rotation matrix + scale are GPU-decoded into half-float rotation/scale textures so the
streamed splats participate in voxel-based IBL shadowing (matching the non-stream path). Standalone: the work
buffer owns the rotation textures. Hosted: the compound's rotation textures become a shared render-target atlas
the stream decodes into. Defaults to false.
Optionalsplat
Enables budget-driven LOD: caps the total rendered splats by converging a screen-space (size + distance)
pixel-size threshold to the budget. Selection is view-direction-independent, so it is consistent across any
number of active cameras (each node takes the finest level and largest projected size any camera demands).
A number is an explicit splat cap; "auto" picks a device-tiered default (desktop 2.5M / iOS 1.5M /
other mobile 1M; XR shares the mobile tier). Undefined (default) disables the cap — LOD is pure
distance, identical to prior behavior. Runtime-mutable via the splatBudget accessor. When this stream
is hosted in a compound, the compound's GaussianSplattingMesh.splatBudget (if set) overrides this and
apportions a shared budget across all its streams.
Options for GaussianSplattingStream.