Raw scans are too heavy to use directly
LAS/LAZ is accurate, but poorly suited for inspection, sharing, and repeated review. The friction to show decision-makers is too high.
LiDAR-to-Gaussian Splatting engine, runs locally
LiDAR to editable Gaussian Splatting, on your laptop.
PointPeek is a local engine that converts LAS/LAZ scans into lightweight, editable 3D Gaussian Splatting assets without cloud uploads or expensive workstations. Survey-grade fidelity, ready for operations.
Local processing benchmark
Turned a 140-million-point road LiDAR dataset into a lightweight, editable asset
with survey-grade fidelity — entirely on a standard laptop.
- Input
- 140 million-point road LiDAR
- Environment
- Standard laptop, fully local
- Output
- Lightweight asset, survey-grade fidelity
Demo gallery
See it working.
Converting a large-scale real-world LiDAR scan of Seokchon Lake in Seoul into an editable 3D Gaussian Splatting asset locally on a laptop.
Demonstrating real-time object segmentation and editing in Gaussian Splatting assets using integrated AI segmentation.
A performance demo processing a 5.5GB LAS dataset of a 3km × 2km urban area, with real-time rendering as both point cloud and Gaussian Splatting.
Converting and reconstructing real-world Paris-CARLA-3D LiDAR data into a 3DGS asset via PointPeek's local engine.
Why it matters
Current point cloud workflows break down at the inspection, update, and sharing stages.
Cloud dependency strains both cost and security
Uploading large spatial data every time accumulates friction in cost, time, and security.
Assets created once cannot be updated
Roads change daily. But traditional Gaussian Splatting is hard to partially modify — even small changes require rebuilding everything from scratch.
The core idea
Survey-grade fidelity, reliable results. And — it's editable.
PointPeek delivers survey-grade fidelity in the Gaussian asset, with consistent, reproducible results you can rely on in production, and output structured for granular scene editing — add, delete, and update. The validation process is straightforward—send us your LAS/LAZ sample, and we will reply with the converted results.
Request sample data verification
01 / convert
Survey-grade conversion
LAS/LAZ scans become Gaussian assets with survey-grade fidelity. Results are consistent and reproducible.
02 / handle
Large-scale local processing
Handle 60GB-scale data on ordinary laptops. No cloud uploads, no external exposure.
03 / edit
Granular scene editing
Add, delete, fill regions, copy and paste. Just re-convert the changed road segment.
Workflow
From field scan to operational 3D asset, on a single machine.
-
1
LAS/LAZ input
Input native LAS/LAZ files directly. No preprocessing, no pre-conversion, and no uploads required.
-
2
Local conversion
Generate Gaussian assets on an ordinary laptop, with consistent, reproducible results every run.
-
3
Inspection and editing
Review point cloud and Gaussian Splatting side by side, and edit (add, delete, copy) on a region-by-region basis.
-
4
Operations and scale
Lightweight assets extend into web sharing, partial updates, and AI and simulation integration. No need to rebuild from scratch.
Built for spatial operations
Fields where reality-based 3D assets need to work at the operational stage.
HD map production and updates
Update only the changed road segments. No need to re-scan the entire city.
Autonomous driving simulation
Repeat QA, route inspection, and scenario simulation in lightweight reality-based environments.
Digital twin operations
Use measured point cloud data as the base for visual and operational twins, accumulating and updating changes.
Physical AI environments
Convert real-world spaces into training environments for robotics and autonomous driving. Reality-based assets that complement the limits of synthetic data.
For Physical AI
The gap synthetic data can't close — filled with real, measured spaces.
Physical AI trains in simulation and must work in the real world. The sim-to-real gap can't be closed with synthetic data alone. PointPeek feeds the training pipeline with actually measured spaces — as editable, updatable, scenario-ready assets.
Environments grounded in measured data
Survey-grade, millimeter fidelity in the Gaussian asset. Not a plausible-looking synthetic scene — actual roads, factories, and urban spaces used directly as training environments.
Build scenarios on top of measured environments
Add vehicles, place pedestrians, create construction zones — on the same road. Starting from a measured base and applying variations only, scenario diversification is fast and consistent.
Environments that accumulate real-world change
When a road changes, update only that segment. The training environment doesn't get built once and frozen — it stays alive by accumulating real-world change.
Training assets with no external exposure
Survey data, factory interiors, secure facilities — data that can't leave the premises is converted locally and used for training. The data never leaves the client machine.
Built for
- Autonomous driving SIL/HIL simulation
- Robotics policy training environments
- Sim-to-Real domain transfer
- Reality base for synthetic data augmentation
About us
PointPeek is a LiDAR-to-Gaussian Splatting engine for operating reality-based 3D data locally.
Built to directly solve problems encountered in real operational environments. First validated on road-based LiDAR, and expanding collaboration to diverse real spatial datasets.
- Adoption and PoC inquiries
- Sample data-based performance validation
- Indoor, urban, or industrial space data collaboration