Team of Silicons was assigned to build a Vision Model for Autonomous Surveillance Drones to detect and flag troops in border dispute areas.
ToS built a multimodal edge vision model, fusing RGB and thermal streams that detects and tracks objects across frames. Instead of one over-confident classifier, pairing a real-time edge detector (distilled) with a dedicated open-world layer that flags when evidence is inconclusive works better.
Alerts land as intel on unverified person, vehicle, unrecognized object, or insufficient evidence for an analyst to judge.