Maritime
Training data for vessels that steer and watch for themselves.
Obstacle detection, scene segmentation and vessel behaviour across optical, thermal and drone imagery, with live review on deployed models.
From the horizon to the harbour wall
Every part of the workflow where we can take work off your team.
Navigation
Surveillance
Port and coast
Live operations
We work with optical, thermal, drone and AIS.
What we label
Every class is written into a guide with your team before the first batch. Files come back as COCO JSON, YOLO, Pascal VOC, CSV, JSONL or your own schema.
Optical and thermal
Vessels and obstacles
Bounding box and track ID
Cargo ship, tug, kayak, buoy, debris
Scene
Segmentation
Water, sky, land, horizon line
Range
Class
Near horizon, mid range, close approach
Drone and aerial
Port and coast
Polygon and bounding box
Berth, crane, container stack, oil sheen
Tracks and AIS
Behaviour
Time span and class
Loitering, rendezvous, dark vessel, speed anomaly
How we test the model
The tests that matter for maritime models, and what you get back from each.
Small target recall
Distant and partly hidden targets reviewed as their own set.
You get backRecall by range and sea state
Sensor agreement
Detections compared across optical and thermal cameras.
You get backDisagreement rate with examples
Weather and light recall
Detections reviewed on purpose in fog, rain, glare and at night.
You get backRecall by condition, failing clips for training
Where maritime teams start
Problems teams bring us, and the first piece of work we would scope.
Small boats near the horizon are invisible to your model until it is too late.
What we doTargets labelled by range and reviewed as a separate set, with misses fed back to you.AnnotateGlare, spray and fog make every frame look different.
What we doLabels across sea states, light and weather, from optical and thermal cameras.AnnotateAn uncertain detection on a moving vessel cannot wait for tomorrow.
What we doA reviewer checks it live before it reaches the operator.Live reviewRelated work
2,000 robot videos described every week by a team built in seven days
Half a million images for autonomous navigation, with every file kept on EU servers
Twelve times the volume in five weeks, with quality holding at 95 to 98%
Human in the loop support for visual deviation detection on production lines, with a response time requirement under two minutes.
24/7 live review for production line visual AI, quality above 95% by month two
Bring us the maritime data your model gets wrong.
Send a small batch and get it back done, free, usually within 24 to 48 hours. You judge the work before you pay for anything.