Designing affordable and privacy-preserving technologies to advance food equity: an on-device shelf-monitoring prototype for the Cal Poly Food Pantry. PI: Prof. Souvik Kundu · Student researcher: Amogh Somisetty.
Start here. The full process in order, including the setbacks: how the site was characterized, how the baseline model was trained, both domain-adaptation experiments, the switch to a second GPU provider and what broke along the way, and the dataset expansion in progress now. Diagrams and explanations throughout.
Pre-semester progress, summarized for the PI: the system concept, the trained baseline model, both experiments with before/after visuals, the groundwork (literature, IRB, privacy architecture), and the five decisions queued for the first meeting.
The deployment site, measured before visiting: dimensioned floor plan and shelf elevation from video photogrammetry, camera coverage geometry, annotated site frames, the trainable-feature inventory, and the monitoring-loop demonstration.