MapGen v3
Aerial-photogrammetry processing daemon by the Atlas Imaging map team.
Install
Deploys the MapGen v3 Docker stack (redis, mapgend) on an Ubuntu 22.04 GPU host. The current release candidate:
curl -fsSL https://mapgen.isaret.ai/install.sh | MAPGEN_CHANNEL=rc sh
The stable channel has no release yet (v3.2.0 will be the first); once it exists, it is the default:
curl -fsSL https://mapgen.isaret.ai/install.sh | sh
The installer pulls the prebuilt image, writes docker-compose.yml and .env (with a generated redis password) into /opt/mapgen, starts the stack and waits for /api/v3/health on port 1307. The API has no authentication: run it on a trusted network. Docker publishes the port around ufw, so ufw deny 1307 does not close it; set MAPGEN_BIND in .env to one interface's address to narrow it. Rerun it to update; your settings stay, and it stops while a flight is running (sh -s -- --force updates anyway and kills the flight). It asks for the key you were given, or reads MAPGEN_DOCKER_USER and MAPGEN_DOCKER_TOKEN. Do not put sudo in front: the script asks for it when it needs it, and sudo would drop the variables. Read the installer first if you like: install.sh. The compose file is not on this site: it ships inside the image, so it always matches the image it starts, and the installer extracts it after the pull. To read it, after docker login: docker run --rm --entrypoint cat atlasimaging/mapgen_v3:rc /usr/share/mapgen-v3/docker-compose.yml. Afterwards it is plain Docker Compose: cd /opt/mapgen && sudo docker compose ps. Stop and remove it with sh -s -- --uninstall (data volumes stay).
What you need
- Ubuntu 22.04 (jammy)
- An NVIDIA GPU (Turing, Ampere or Ada) and driver R550 or newer
- Docker Engine with the Compose plugin, and the NVIDIA Container Toolkit
- The key: a Docker Hub access token for the
atlasimaging/mapgen_v3image, and its account name
The installer checks all of these first and tells you what to install; it never installs Docker or NVIDIA packages itself.
Links
- GitLab project (team access)