Computer vision on edge devices
summary
Built a working self-checkout vision pipeline that captures camera frames tied to receipt logs, trains on 1,000,000+ annotated photos, and runs real-time inference on Nvidia AGX edge inference units to flag mismatches and suggest product choices. Cheap off‑the‑shelf cameras require reduced shutter time and brightness tuning; initial models distinguish similar items (bananas, beans vs peas) using a frame‑vote consensus and face scaling limits with class‑based nets, so the team pivots to vector + catalog nearest‑neighbor lookup to support tens of thousands of SKUs.
AI-written summary · about
related talks
27mComputer vision for automated quality inspection
57mWhat's that fruit?
41mPractical Aspects of Applying Modern Computer Vision Technologies
20mComputer Vision at the Edge with OpenVINO
36mAI-Powered Real-Time Quality Control
31mThe Rise of No-Code AI Vision at the Edge
09mLightning Talk: Streaming and Processing Edge Vision Data in Real Time
06mAI in production: The droids you’re looking for