Production and Conversion on Your Own Computer with CartaX Studio

Not every job needs the cloud. If you have a suitable computer, CartaX Studio produces the same products — orthophoto, DSM/DTM, point cloud — on your own machine, converts your LAS/LAZ and GeoTIFF files into the format the browser opens, and uploads only the result to your project. No tokens are spent; no 40 GB of raw data is uploaded. This guide walks through the local route; for what Studio is and the installation details see What is CartaX Studio?.

Installation: Windows and Linux

• Windows 10 / 11. Get the installer with Download on the CartaX Studio card of the dashboard. Because the file is not signed yet, Windows shows "Windows protected your PC"; that is normal: More info → Run anyway. Nothing else is needed on Windows 11; on Windows 10 the Microsoft Edge WebView2 runtime must be installed.
• Linux. Download and install the .deb package.
• There is no macOS build.

Studio is part of the Professional plan; a free account cannot sign in. At sign-in Studio opens your browser; once you sign in with your CartaX account you return to the app.

The production engines are not part of the installer; they are downloaded once, on your first production (about 0.4 GB). Conversion needs no engine download.

Conversion: LAS/LAZ and GeoTIFF

Press Convert and pick the file; Studio detects the format itself.

Input	Output	What you get in CartaX
LAS / LAZ point cloud	Ready-made octree	Smooth 3D viewing in the browser, measurement, cross-sections
GeoTIFF orthophoto / DEM	COG	A georeferenced raster layer on the map, elevation readout

When conversion finishes the result is uploaded to the project you chose automatically. Uploading the same file to the cloud would cost tokens; local conversion does not.

Production: from a photo folder to products

The Create button opens the production wizard:

1. Pick the photo folder; subfolders are scanned too and the photo count appears immediately.
2. Tick the outputs: Orthophoto, DEM (DSM/DTM), Point cloud. Under the point cloud there is an optional Classification — points are split into ground, vegetation, building, road and vehicle classes. The Textured mesh (3D Tiles) output currently shows a coming soon badge.
3. Add a GCP file if you have one; you can mark the points on the photos and get an accuracy report at the end of the job.
4. Choose the delivery coordinate system (Automatic is recommended).
5. Press Produce on this computer.

Production runs at a single quality, High. Before the run the wizard measures free disk and memory; if they are not enough for this job the local button is disabled, hovering it shows the reason, and the same job can be sent with Produce in the cloud in one click.

No tokens spent, no raw data leaves

Local conversion and local production spend no tokens. Your raw photos and raw LAZ never leave your computer; only the produced outputs are uploaded to the project and count against your storage quota. The outputs are also written on your computer under Documents / CartaX Studio in product folders: Orthophoto, DEM, Point Cloud, Mesh, Report. The Report folder holds the run log and, if any, the accuracy report — your delivery files are in your hands.

The Jobs screen: progress, Pause / Resume, Retry upload

The Jobs screen shows overall progress on a single card; open it to see each job's stages (photogrammetry, classification, conversion, upload), elapsed time and percentage. You can queue several jobs; processing runs one at a time, uploads run in parallel.

• Pause / Resume — suspend a running job instantly to free the computer for other work; it continues where it left off.
• Safe across restarts — if Studio closes or the computer restarts, the job resumes from the stage it reached at the next launch.
• Retry upload — if the upload fails on a network error, the products are ready on your computer; this button repeats only the upload to the web, production is not re-run.

While the upload is in flight the CartaX Studio card in the web app shows the job as running; once it completes the layer opens on the map. If you ticked the DEM, the Measurement elevation: Surface / DSM / DTM selector appears in the layer settings.

Duration and machine power

Photogrammetry is the heaviest step: hundreds of photos can take hours. Studio uses all cores; a machine with an NVIDIA graphics card is noticeably faster. To shorten the run you can drop classification.

There is no hard cap on the number of photos; your machine's memory and disk space decide. The upload dialog on the web reminds you that sets of up to 800 photos can be produced for free with Studio on a suitable computer; for larger sets trust the wizard's disk/memory check — if it is not enough, the button is disabled anyway.

Minimum system requirements for segmentation

Segmentation recognises the objects in your photos (buildings, roads, vegetation, vehicles, wires, people…) and feeds that knowledge into the rest of the run: point cloud classes, the class layers that appear on the map, and the masking of people in the photos all come from it. From version 0.6 it runs on this computer, with the same model and the same version as the cloud.

One thing sets it apart from the other outputs: it runs on the graphics card and there is no CPU fallback.

Requirement	Minimum	Why
Graphics card	NVIDIA (CUDA-capable)	The model only runs on NVIDIA cards. AMD, Intel and Apple graphics, and integrated GPUs, are not supported.
Graphics memory	8 GB	Measured peak usage is 5.6 GB; the remainder covers what your display itself uses. On 6 GB cards the job stops with an out-of-memory error.
Free disk space	14 GB	The engine downloads once at 5.5 GB and unpacks to 7.3 GB; both sit on disk together for a while.
System memory (RAM)	16 GB	The minimum Studio already asks for on every production job.

A few notes:

• There is a one-time ~6 GB download on first use (Settings → Production engines). It is not part of the installer; later jobs do not download again.
• Your photo resolution does not affect graphics memory. The model scales each image to a fixed size, so a 12 MP set and a 45 MP set use the same memory. Only the card's total memory matters.
• Duration scales with photo count: measured at ~26 seconds per frame on a cloud-class card. A 200-photo set ≈ 1.5 hours, on top of photogrammetry.
• Selecting segmentation also produces the point cloud. The class layers are derived from the classified point cloud, so Studio adds it for you.

If your computer does not qualify: Studio will not start segmentation and will say why (for example, "at least 8 GB of graphics memory is required; this computer has 6 GB", or "the segmentation engine is not installed"). You can produce it in the cloud instead — no hardware requirement there, billed with tokens. Orthophoto, DSM/DTM and point cloud outputs are unaffected by these requirements; they still run locally on computers without a graphics card.

Compared with the cloud

	Studio (local)	Cloud
Cost	No tokens	Tokens (by megapixels)
Your computer	Busy for the whole run	Stays free
Classes (object recognition)	Yes (NVIDIA card required)	Yes
Photo cap	No hard cap; memory and disk decide	2,000 per job
Raw data	Stays on your computer	Uploaded to CartaX

For the cloud route see Cloud photogrammetry production.

Related guides

• What is CartaX Studio?
• Cloud photogrammetry production
• Classes: automatic object recognition
• What are LAS and LAZ files?
• Supported formats and limits

Frequently asked questions

Question	Answer
Does it spend tokens?	No; local conversion and production are free, only the uploaded output counts against storage.
Are my raw photos uploaded?	No; they stay on your computer.
Windows says it "protected your PC"?	Normal: More info → Run anyway.
Is there a macOS version?	No; Windows 10/11 and Linux.
How many photos can I process?	No hard cap; memory and disk decide.
Can I pause a job?	Yes: Pause / Resume; after a restart it continues from the stage it reached.
The upload dropped — does production restart?	No; Retry upload repeats only the upload.
Do I get Classes here too?	Yes, since 0.6 — it needs an NVIDIA card with 8 GB of graphics memory (see the table above). Computers that do not qualify can produce it in the cloud.