OpenUSD Point Cloud Plugins

OpenUSD FileFormat Plugins and libraries for point-cloud data. The project opens LAS, LAZ, local or resolver-backed COPC, and PLY sources through shared point-cloud contracts for surveying, mapping, scanning, and 3D data-exchange workflows.

What it does

• Reads LAS 1.2-1.4 and LAZ point records through format-specific plugins.
• Reads local and resolver-backed COPC metadata, hierarchy, and selected byte ranges without coupling the reader to HTTP.
• Reads scalar PLY vertex properties through ASCII and binary point streams.
• Authors UsdGeomPoints, metadata, OpenUSD 26.08 usdLod roots, and payload-backed spatial tiles through shared authoring.
• Supports metadata-only inspection, bounded-memory tiled generation, stable LASxxx / LAZxxx / COPCxxx diagnostics, and explicit conversion for long-running tiled jobs.

Rendering is the consuming application's responsibility. These plugins provide import and USD authoring, not a point-cloud renderer, and LOD selection stays with the host application. See the tile and LOD contract.

Supported Formats

Extension	Plugin	Current support
.las	pointcloud-las	LAS 1.2-1.4, point formats 0-10, CRS, waveform metadata, and Extra Bytes
.laz	pointcloud-laz	Compressed point formats supported by the bundled laz-perf adapter
.copc	pointcloud-copc	Local or resolver-backed metadata-only, direct, and native-hierarchy tiled reads
.ply	pointcloud-ply	Scalar vertex reads, explicit CRS, bounded streaming, and payload-backed tiled reads

Point formats 4 and 5 require LAS 1.3 or newer; formats 6-10 require LAS 1.4. The LAZ adapter rejects waveform formats because the bundled laz-perf codec does not provide their compressed record decoders.

Attribute	Status
XYZ, intensity, return number, number of returns, classification	Authored
RGB (formats 2, 3, 7, 8, 10)	Authored
GPS time (formats 1, 3, 6-10)	Authored
NIR, scan angle, user data, point source ID, classification flags, scanner channel, scan direction, edge of flight line	Authored
Waveform packet metadata and external .wdp reference (LAS formats 4, 5, 9, 10)	Authored
Extra Bytes point attributes	Types 1-30 authored as scalar/vector USD attributes

Extra Bytes are authored as geo:<normalized descriptor name> with the descriptor scale and offset applied. Scalar types 1-10 become double[], vector types 11-20 become double2[], and vector types 21-30 become double3[]. Non-finite values and integer values that are not exactly representable as double are rejected. Names are normalized deterministically to ASCII USD identifiers; original names remain in header metadata. See the capability matrix.

The complete matrix, including VLR, CRS, and authored USD attributes, is in the capability matrix.

The capability matrix is the canonical source for point formats, attributes, CRS, metadata, and authored USD details. Source identity, caching, and adaptive tiling are implemented. The next milestones measure that infrastructure on real data, unify tile planning across sequential formats and the COPC native hierarchy, and mature resolver-backed source identity before E57 or other format expansion. Terrain, raster, and vector formats are future repository candidates.

Quick Start

Requirements: CMake 3.23+, a C++17 compiler, OpenUSD 26.08 (the plugin contract accepts >=26.08,<27.0), and OpenStrata 0.21.0 for the pinned workspace build.

Build and test everything through the pinned OpenStrata workspace:

ost configure
ost build
ost test

Build and verify a single bundle:

ost plugin build .\plugins\pointcloud-las
ost plugin test .\plugins\pointcloud-las --up-to 4

Generate a tiled, payload-backed asset explicitly for production processing:

usd-pointcloud-convert `
  C:\path\to\sample.las `
  C:\path\to\output\PointCloud.usda `
  --tile-size 128 `
  --memory-limit 1048576

The converter is the operational path for long-running tiled generation. It publishes the root layer and a deterministic PointCloud.usda.manifest sidecar after generation completes; static FileFormat tiled reads remain available for preview and small inputs. The sidecar records normalized generation arguments and relative payload asset paths. The converter can reuse committed deterministic USDC cache entries through --cache-root; direct LAS, LAZ, COPC, and PLY FileFormat cache lookup uses USDGEO_CACHE_ROOT.

Build and test the libraries with plain CMake — without an OpenUSD runtime, this covers usdGeoCore, usdPointCloudCore, usdLas, and usdLaz:

cmake -S . -B build -DUSDGEO_BUILD_TESTS=ON
cmake --build build --config Release
ctest --test-dir build -C Release --output-on-failure

The full surface is in BUILDING.md.

Using the Plugins

Point PXR_PLUGINPATH_NAME at the directory holding plugInfo.json. A trailing slash makes OpenUSD search subdirectories, registering both bundles at once:

$env:PXR_PLUGINPATH_NAME = "C:\path\to\pointcloud-las\plugin\resources\"

export PXR_PLUGINPATH_NAME=/path/to/pointcloud-las/plugin/resources/

Then open a file with any OpenUSD tool, or reference it from a layer:

usdview sample.las
usdcat --flatten sample.las -o sample.usda

def "Survey" (
    references = @./sample.las:SDF_FORMAT_ARGS:lod=balanced@
)
{
}

The supported compact profiles are off, preview, balanced, and quality. tile=true connects the LAS/LAZ/COPC/PLY stream to spill-backed, payload-backed spatial tiling. tileSize, tileMemoryLimit, and payloadDirectory control the tiled output. Registration details are in INSTALL.md; the full argument surface is in the file-format argument contract.

Previewing without installing

ost plugin view prepares the managed OpenUSD runtime and the plugin environment for a bundle, so a built bundle can be inspected in place:

ost plugin view `
  .\plugins\pointcloud-las `
  C:\path\to\sample.las `
  --with .\plugins\pointcloud-laz

ost plugin view \
  ./plugins/pointcloud-las \
  /path/to/sample.las \
  --with ./plugins/pointcloud-laz

--with makes one session discover both LAS and LAZ. Compact LOD profiles are passed in the standard USD format-argument suffix:

ost plugin view `
  .\plugins\pointcloud-las `
  'C:\path\to\sample.las:SDF_FORMAT_ARGS:lod=balanced' `
  --with .\plugins\pointcloud-laz

To run any other USD tool under the same composed environment:

ost plugin run .\plugins\pointcloud-las -- usdcat --flatten sample.las -o sample.usda

Authored result

#usda 1.0
(
    metersPerUnit = 1
    upAxis = "Y"
)

def Points "PointCloud"
{
    point3f[] points = [(0, 0, 0), (1.42, 0.31, -2.07), ...]
    custom uchar[] geo:classification = [2, 2, 6, ...]
    custom int[] geo:intensity = [1024, 981, ...]
    custom uchar[] geo:numberOfReturns = [1, 2, 2, ...]
    custom uchar[] geo:returnNumber = [1, 1, 2, ...]
    custom double3 geo:boundsMin = (0, 0, -84.31)
    custom double3 geo:boundsMax = (512.5, 43.2, 0)
    custom double3 geo:localOrigin = (-8242.5, -34212.5, 12.4)
    custom string geo:linearUnit = "metre"
    custom string geo:upAxis = "Y"
    custom int geo:epsgCode = 0
    custom string geo:wkt = "PROJCS[...]"
    custom uint64 geo:pointCount = 1048576
}

Positions are stage-local float values. Source coordinates are recovered from geo:localOrigin; the exact expression is in the capability matrix.

Tiling and LOD Status

Library capability and FileFormat integration are deliberately reported separately, because they differ today.

Authoring library (usdPointCloudAuthoring):

• usdLod hierarchy authoring: implemented
• spatial tiled authoring: implemented
• payload-backed tile output: implemented

LAS/LAZ/COPC/PLY FileFormat read path:

• compact non-spatial LOD profiles (lod=preview|balanced|quality): implemented
• spatial tile argument: implemented
• bounded-memory payload generation during file open: implemented
• PLY scalar point reads and payload-backed fixed-grid tiling: implemented

The read path consumes bounded pull-stream chunks, spools points by source-coordinate tile, and authors one payload-backed level per tile. A Windows baseline against the supplied 14.6-million-point Shizuoka LAS input is recorded in streaming and tiling; payload working-set measurements are included, while broader real-world dataset coverage remains open.

Known Limitations

• Tiled reads use bounded chunk delivery and spill points to per-tile temporary files. A single Shizuoka LAS RSS/spool/payload baseline is published in the streaming and tiling roadmap; broader real-world dataset coverage remains open.
• File-format arguments expose normalized attribute selection, chunked and range-based reads, compact lod profiles, spatial tiling, bounds filters, and classification filters. The two filters are evaluated in source coordinates before stage-local transforms.
• metadataOnly reads author the /PointCloud metadata namespace without decoding point records: source count, bounds, CRS, and available-attribute metadata, but no point positions.
• Extra Bytes support covers types 1-30. Non-finite values and integers not exactly representable as double are rejected; descriptor names are normalized deterministically before USD authoring.
• CRS WKT is retained and EPSG is inferred from explicit WKT or GeoTIFF horizontal CRS keys. Conflicting definitions are rejected with a typed ConflictingCrs diagnostic.
• The deterministic USDC cache is available to the conversion tool through --cache-root; direct FileFormat lookup reuses committed entries through USDGEO_CACHE_ROOT.
• Writing LAS, LAZ, or COPC is out of scope; all three plugins export as usda.

See the implementation status and roadmap for the planned work.

Status

Latest release: v0.8.0 — real-world fixed-grid and adaptive measurements, I/O observability, and LAZ point-format-7 hardening. The v0.3.0 module and bundle rename is recorded in MIGRATION.md. See the release record and CHANGELOG.md.

Direction is fixed in the design policy; the structure is fixed in the workspace contract.

Architecture

.las  -> usdLas  ----\
.laz  -> usdLaz  -----+--> usdPointCloudAuthoring --> pointcloud plugins --> stage
.copc -> usdCopc -----/
          ^
          usdGeoCore + usdPointCloudCore
          (shared geospatial and point-cloud contracts)

The repository separates format readers, shared validation, USD authoring, and plugin adapters so the core tests remain independent of an OpenUSD runtime. The usdPointCloudTiling module provides format-independent tile routing and spool contracts for the streaming authoring path. Readers remain independent of OpenUSD and the tiling library remains independent of LAS and LAZ; the plugin adapters connect those pieces through shared APIs. The binding version of this is WORKSPACE.md.

Repository Layout

libs/usd-geo-core/              Geospatial values, transforms, bounds, cache keys, diagnostics
libs/usd-pointcloud-core/       Point attribute, chunk, read-option, sampling, and LOD contracts
libs/usd-pointcloud-authoring/  OpenUSD point-cloud, usdLod, and payload authoring
libs/usd-las/                   LAS header, metadata, and point-record reader
libs/usd-laz/                   LAZ chunk reader and laz-perf adapter
libs/usd-copc/                  COPC metadata, hierarchy, and local range reader
libs/usd-ply/                   PLY 1.0 header-inspection foundation
plugins/pointcloud-las/         LAS OpenUSD FileFormat Plugin
plugins/pointcloud-laz/         LAZ OpenUSD FileFormat Plugin
plugins/pointcloud-copc/        COPC OpenUSD FileFormat Plugin
docs/                           See docs/README.md for the documentation index

Every module under libs/ and plugins/ carries its own README.md stating what it owns and refuses to own; that README is part of the module contract. See the module README contract.

Documentation

Start at the documentation index.

• Workspace contract — structure and dependency directions
• Capability matrix — what is supported today
• Building and testing · Installing · Binary distribution and licensing
• Design policy
• Tile and LOD contract · Plugin adapter · File-format arguments · Point reader · Diagnostics
• Roadmap · Streaming and tiling · Implementation status
• OpenUSD compatibility · Migration
• Release records · Third-party notices

License

Project code is licensed under the Apache License 2.0; see LICENSE and NOTICE. The LAZ adapter incorporates laz-perf 2.0.0, which is distributed under LGPL-2.1, and the pointcloud-laz plugin binary therefore contains LGPL-2.1 code. Redistributing pointcloud-laz binaries carries the obligations described in binary distribution; see THIRD_PARTY_NOTICES.md for the applicable third-party terms. The pointcloud-las plugin contains no laz-perf code.