Why AIDroneMap

The problem

Drones capture more images, at higher resolution, than ever before. A normal professional survey is a few hundred images of 20 megapixels or more, several gigabytes of data. Turning them into orthophotos, elevation models, point clouds and 3D models takes serious computing power, and that power costs money.

Today the usual choice is a machine big enough for your largest job that sits idle most of the time, or a long wait.

Hillshade of the vineyard surface model: the vine rows read as fine parallel ridges, with taller, rougher texture along the treelines and two bright access tracks. — Hillshade of the DSM. aidm-vineyard-small, 73 images, processed by AIDM 0.1.0, dataset commit `6af198c9`

The idea

Compute you rent by the second, or already own, joined into one pool for as long as a job needs it and released when the job ends. One job, spread across every CPU and GPU you can reach, started with one command.

The engine is the base of that. The cloud layer, the next piece, is the scheduler that spreads the work.

What we build for

AIDM is built for large datasets and modern machines. A normal job is 500 images and 3 GB of data. The engine targets current CPUs and GPUs and does not keep compatibility with hardware that cannot run such a job. The exact minimum CPU is published with each release on aidronemap.dev.

Principles

Simple to deploy, simple to use. One engine image, one command to run it, one URL to point a client at.

Keep your tools. A compatible API and familiar outputs, so nothing you use today has to change.

Pay only while it runs. No machine left idle, and no single cloud provider required.

Measured in public. Every speed or accuracy claim comes with a benchmark against ODX, with the method and the datasets published. This release makes no such claim; the first numbers come in January 2027.

Open and independent. The engine is licensed AGPL-3.0, and there are no paid features in it. Sponsorship never buys features.

Where it comes from

AIDM is a fork of ODX, the engine behind WebODM, which continues the work of OpenDroneMap. AIDroneMap exists to take that engine in a direction its maintainers are not focused on: distributed processing on rented compute, and automation with AI in the loop. It is an independent project, not affiliated with or endorsed by WebODM or OpenDroneMap, and ODX remains the reference for production work today.