A pipeline that generates synthetic EO/IR imagery already labeled with perfect bounding boxes, so a perception team can bootstrap a detector without scarce, expensive real data.
What you're looking at
The grid: synthetic images across five target classes (ground vehicle, light truck, small vessel, fixed-wing UAS, rotary UAS). Every box was generated automatically, because we placed the target, we know exactly where it is. No human labeling.
EO vs. IR: color (electro-optical) and a simulated thermal (IR) rendering, with hot-spots where engines/exhaust would be.
Filters & stats: slice by class, modality, terrain, or noise; the bars show class balance.
Try this
Filter to IR and notice the boxes still sit exactly on the targets.
On your engagement
We build the pipeline around your target classes and sensors, exported in your training format (YOLO / COCO).
Real vs. demonstration. This is a methodology demonstration, not training-ready imagery. We measure and state the sim-to-real "domain gap" honestly (currently MEDIUM) and recommend validating on 15 to 20% real data.
GIDEON CORP.