Home / Industries
IndustriesAI data for the domains where mistakes cost most.
The same image can be labelled six different ways, and only one of them fits the system you are building. We set guidelines, processing and review around the context your model will operate in: the sensors, the classes, the edge cases and the cost of an error.
Six domains, one method.
Select a domain to see an annotated example, marked up the way that domain’s data is usually structured.
Aerospace, medical AI and geospatial.
These domains run on the same managed workflow as the pages above. What changes is the data problem, the inputs and what a reviewer has to check.
Airborne and space-based imagery
Data problem. The same object looks different from every altitude, sensor and sun angle, so a class definition written for one collection rarely holds for the next.
Typical inputs. Airborne EO, satellite imagery, and EO/IR frames from aircraft or drones.
Protocol-led imaging annotation
Data problem. Labels are only useful if every reader applies the same protocol to the same structure, slice after slice and study after study.
Typical inputs. DICOM studies.
Tasks. Segmentation of anatomical structures, outlines and measurements, and validation of model outputs against a written protocol, reviewed in the browser in Point AI Studio.
Drone, satellite and multi-sensor imagery
Data problem. Dense scenes with hundreds of small objects per tile, at mixed scales and orientations, over terrain that changes between collections.
Typical inputs. Satellite and drone imagery, overhead video and multi-sensor captures.
Tasks. Oriented boxes on vehicles and structures, segmentation of regions and footprints, and object tracking in overhead video.
Data handling. Work is done by Point AI’s employed, in-house team in Tel Aviv and Haifa. We do not list certifications or regulatory approvals here; if your programme has specific requirements, raise them in the first conversation.
Have a challenging data problem?
Tell us what you’re building and what data you have. We’ll suggest a practical way to process, check and deliver it.