OpenLiDARViewerGUIDE · v0.6.9
This guide walks through opening a scan, finding your way around, measuring, analysing terrain, comparing two scans and sharing your work. It assumes no GIS background. Where the viewer refuses to do something, the guide says why.
The moment a scan lands you get a one-line summary and the most useful next step as a single button. Nothing runs until you ask.
Streaming. A .copc.laz or an ept.json, on disk or at a CORS-enabled URL, opens through progressive octree streaming: partial range reads, a view-dependent scheduler, bounded residency and worker decoding. A 3D Tiles tileset.json whose content is PNTS streams the same way, explicit or implicit hierarchy, ADD or REPLACE refinement. Paste the URL into the start screen or share a ?copc=<url> deep link.
A local LAZ opens progressively. NEW IN 0.6.9 The header prefix is read first, a worker frames the file from it, and chunks decode from a pool behind a strided stand-in preview, so the compressed file is never held whole. The preview is a sample, capped at the smaller of the render budget and two million points, and is marked incomplete until the decode finishes. An ordinary LAS still reads whole under the memory guard.
Heavy local files. A very large uncompressed LAS or a chunked LAZ is indexed into temporary browser storage and streamed from there instead of being read whole. A preview sample renders at once, marked incomplete until the full index swaps in. Reopening the same file reuses the index. If the device cannot hold even the index, the file is refused with guidance rather than crashing the tab.
RGB the scan’s own colour. Height low to high, percentile-clipped 5/95 on a Viridis ramp. Intensity reflection strength. Class by ASPRS code. Density for coverage QA. Normals. Coverage and Confidence describe the terrain analysis and stay grey until you run it.
A screen-space depth cue that darkens discontinuities so edges, ridges and near/far structure become legible. Presets Subtle, Balanced and Inspection. On by default on desktop WebGPU, off by default on WebGL 2 and mobile, always switchable. It is not physically based lighting.
Chip rails in the Inspector for colour mode, RGB preset (Natural, Photoreal, Drone RGB, Mobile LiDAR, Survey, High Contrast, Infrastructure), depth and background. White balance with an Auto assist on streaming COPC. Soft splats or Inspection radius. A patch view projects a point’s neighbourhood onto its tangent plane.
One Speed to Quality control in the top bar drives streaming preset, resident point budget, concurrent decodes, pixel-ratio ceiling, EDL and antialiasing. It never touches the static load budget that terrain analysis, measurement and export read, so no slider position changes how a number is computed. On a streamed scan it does set the resident point budget, so it can change how much of the source a resident-only result was computed over. The result says so when that happens.
Start Measure from the Tools tab or press M, then pick a kind on the toolbar over the view and click points on the scan. Snapping is off until you turn it on: the snap button cycles off, point and geometry, and it needs points in memory, so a streaming scan has nothing to snap to. Every measurement is editable, undoable and switchable between metric and imperial.
Straight line between two points.
Total length of a multi-segment path.
Polygon area, both true in-plane and horizontal map-projected.
Vertical difference between two points, in the source vertical unit.
The angle at a vertex between two arms.
Rise, run, slope angle and grade percent between two points.
Cross-section between two points with a resizable height-vs-distance chart, validated at E4 against OGR and R.
Cut and fill against a base level from a polygon or 3D lasso, with degenerate and self-intersection guards. A lasso stockpile is integrated over a polygon-clipped grid rather than summed over the points that happen to be loaded. NEW IN 0.6.9 That area-weighted figure is live in the toast only; the session record, the report and the CSV still carry the older point-sample volume, so the two can disagree on the same lasso.
The trust badge. A number is only as good as the points under its ends. A measurement on a fully loaded scan carries a red, yellow or green dot graded on whether its endpoints landed on real returns and how dense the surrounding data is. A streaming scan holds no resident points to grade against, so measurements taken on one carry no dot. Hover the dot to read exactly why. The badge travels with the measurement into the session file, so whoever opens it sees the same verdict.
Measurement chains combine placed measurements as sum, difference or ratio. The whole set exports to a JSON session and re-imports later.
Open the Analyse panel and run the analysis. The viewer classifies the ground, builds a gridded DTM with per-cell confidence and hold-out RMSE, derives the surface products, and issues one Terrain Assessment verdict. A scan that only measured part of the ground says so rather than pretending the filled-in areas are survey quality.
Contour Studio turns an analysed scan into a deliverable. Pick a purpose (Engineering Plan, Survey Review, Terrain Research, Presentation Map, Custom); a purpose only bundles presentation defaults and can never raise a claim. Analytical contours are the exact isolines of the grid. Cartographic contours are generalised for legibility, reference the analytical geometry’s hash and are never labelled exact. An opt-in terrain-aware mode keeps more shape on low-confidence support and small closed features; the simplifier still refuses to drop a gap or a low-confidence vertex.
Contours also draw over the point cloud as a derived layer with visibility, index emphasis, opacity and vertical lift. A segment crossing a data gap is not drawn, so a line never spans unmeasured ground.
An analysis is kept once it is computed. NEW IN 0.6.9 The terrain core is written to origin-private storage on your own disk and restored when you ask for the same analysis again, which turns a six-to-nine second recompute into a restore of well under a tenth of a second, with byte-identical grids. The key is the analysed input rather than the file, so a different stride or a different classification recomputes. Only cores that took a second or more are kept, the store is capped at 512 MiB by least-recent use, and the browser may clear it at any time.
NEW IN 0.6.9 When a scan opens, the Analyse panel carries a Dataset Story: what the scan is, what limits it most, what it is good for, what to treat with caution and what is not established. It synthesises what the other panels already found; it never upgrades a verdict.
NEW IN 0.6.9 Four numeric paths in the analysis were rewritten for speed in v0.6.9 and held to byte-identical output against the code they replace, so the surface is the same one, computed faster.
Process Studio in the left rail lists what the loaded scan can safely produce, each as ready, review or blocked with a plain reason and the shortest action that lifts it. It reads the readiness services. A contour or DEM file is decided by a finer gate that also weighs the evidence grade and the precision permit, so a product can read ready here and still be refused at the moment of export, with the reason given there.
Open a second scan of the same place and the viewer offers a comparison. Two georeferenced layers that declare the same projected CRS mount into one shared frame at their real separation, non-destructively, in Float64.
Two epochs are aligned by a rigid ICP whose default model is yaw plus horizontal translation, so the fit cannot absorb the vertical signal it is meant to measure. The full model is optional and residuals are reported. Epoch alignment sits at E2, and cross-datum comparison is refused rather than approximated.
Per-cell change and the threshold-applied gain and loss volumes are validated at E4 against GRASS over eight synthetic epoch pairs. The raw net volume, summed over every comparable cell with no threshold, is not part of that check. A level of detection keeps changes below the survey’s own precision from being reported as real movement.
The volume’s band is built from the level of detection and co-registration RMSE you supply. With no error budget it reads unquantified, and the compare panel names what to supply, rather than grading a ±0 band as high confidence.
NEW IN 0.6.9 The Tools tab opens on a launcher listing Measure, Inspect point, Annotate and Clip box, each with its key and what it does, and a count of how many you have placed this session. It folds to a strip once a tool panel is up. Probe stays on the dock.
Draw a box and hide everything outside it, or inside it for a cut-away. Type the six extents, or press Fit to scan and work inwards. A point-cloud export made with the converter's Export button carries the clipped points; the quick format buttons write the whole loaded cloud.
Derive ground, vegetation, building and more where a scan carries none. On a scan that already carries producer classes it declines and says so, and Fill unclassified points handles the gaps instead; neither runs on a streaming scan. Return-number aware, with a ground-support gate for roofs and a wall-rescue pass. A derived class is a heuristic guess and is labelled as derived. Paint a class over a lasso and write it back to LAS.
A terrestrial scan arrives as a grid. When the file carries it, click a cell to highlight the point it produced, and back. A cell with no return means the scanner looked and nothing came back. If the scan was reduced to fit in memory the grid survives but the link to individual points does not, and the workbench says so rather than pointing at the nearest point.
Open several scans as layers, group them with visibility, solo and rename, and share one project-wide elevation colour scale so two scans stop stretching their own ranges.
A live summary tells you what each export will contain before you commit. Terrain deliverables decline above a 10 mm coordinate step rather than exporting finer than the representation can distinguish.
Re-save the points as LAS 1.2 or 1.4, PLY, OBJ, XYZ or CSV, respecting any clip and any painted classification. PNG snapshots burn in placed measurements and annotations as inspection evidence.
Multi-page PDF reports from three templates (Survey Summary, Technical Report, Scan QA). Contours as a package with the DXF, a validation record and the Contour Studio settings. Map-sheet PDF, DEM package whose DTM surface carries its own SHA-256, terrain intelligence report, Scan QA report, and a ZIP with a SHA256SUMS manifest. Orthographic RGB, height, intensity, classification, depth and normal images. Measurements as GeoJSON, CSV or KML.
Exported products record which method version produced them, for example olv.contour.generalize.terrain-adaptive@1, and carries the permit decision the gate made.
Filter by source type. Streaming means progressive, bounded-memory loading rather than a full-file read. Support varies with browser memory, GPU capacity and dataset size.
Export: LAS 1.2 / 1.4, PLY, OBJ, XYZ, CSV, PNG. USDZ needs conversion first. Scanner and app compatibility notes live in the full format matrix ↗.
Full GIS layers and analysis, photogrammetry, survey-grade measurement, CRS reprojection and editing of point data are deliberately left to dedicated tools.
The viewer detects the CRS, vertical datum and units, converts known linear units to metres, recentres large coordinates and records everything in export metadata. It never reprojects between systems or transforms vertical datums. Cross-datum comparison is blocked rather than approximated.
Coordinates are recentred before narrowing, so a normal survey keeps sub-millimetre resolution. The step grows with extent: under 1 mm to 16 km, about 8 mm at 100 km. The Scan Report states the figure and terrain deliverables decline above a 10 mm step.
A cloud above the point budget loads at reduced density. LAZ stride loading still decompresses every record because the decoder is sequential. COPC, EPT and the out-of-core path are the routes for very large data.
Its output is derived, not survey-grade. The register grades it on synthetic known-truth fixtures. Separately, agreement against external reference labels on real airborne scenes (OpenGF expert labels, ASPRS producer class-2 ground) is reported with its weak regimes intact: mountain F1 near 0.5, dense-canopy precision near 0.34, building roofs read as ground. Those labels are themselves a classifier’s output, not survey checkpoints, which is why the ground filter stays partial.
If a worker fails, the analysis falls back to the main thread only for sizes measured as safe: 25,000 points and 1,000 estimated cells for terrain, 1,000,000 points for classification. Anything larger refuses with an error that says nothing was produced. Those budgets are measured on one machine as a proxy, not read from your device.
Seventeen claims agree with an independent implementation on frozen fixtures. None has passed a prospective field study with a frozen protocol and independent operators. The app says preview or review wherever that gap matters.
Go to app.openlidarviewer.org and drag a .las, .laz, .e57 or .copc.laz onto the page, or paste a remote COPC URL.
Press R to frame the scan, then 1, 2 or 3 for Orbit, Walk or Fly. Drag to orbit; in Walk and Fly, click the view to take the camera.
Open Scan Intelligence for count, density, spacing and the integrity report. Switch colour to Height or Intensity.
Pick Distance in Measurements, click two points, hover the badge to read its grade.
Run Analyse, read the Terrain Assessment, then open Output. Save an .olvsession to keep the evidence.
An endpoint is not on real data. Move it onto the scan, or turn on snapping.
The scan has no coordinate system, so the viewer cannot confirm the scale. Measurements still work but are flagged yellow, and the terrain rows that depend on a known unit are held back for the same reason.
Run the terrain analysis first. Those modes describe its result.
Open the matching scan file too. The session carries the analysis and the measurements, not the points, and the viewer tells you which file to drop.
COPC, EPT and out-of-core files stream detail-first, so the view sharpens as you look around. A local LAZ opens progressively behind a strided stand-in preview. A preview sample is marked incomplete until the full index is in. A streaming source counts as complete only once its hierarchy is, not once everything it currently knows about has arrived, so an EPT set that opens shallow and deepens reports an honest unknown rather than a premature yes.
No. Files are read and rendered in your browser. There is no account, no server and no telemetry of your data. Remote COPC, EPT and 3D Tiles are fetched by range request from the host you name, and nothing is sent back.
A modern Chromium browser with hardware acceleration, WebGPU where available. Firefox and Safari run through the WebGL 2 fallback and run the full deterministic suite on every release, though Chromium is the engine the release gate blocks on. Phones and tablets are supported with a touch model and a one-handed dock.
Yes. Every archived release has a Zenodo version DOI, and the concept DOI 10.5281/zenodo.21544619 resolves to the newest deposit, which is v0.6.8. v0.6.9 is a release candidate and is not deposited yet. A CITATION.cff ships in the repository. See the science page.