Computer Vision · Mar 2026 - Present
Molt
Molt replaces twice-daily manual poultry headcounts with continuous IP-camera monitoring, using DFCCNet density regression, automated threshold alerts and SMS and in-app notifications.
Stack
- React Native
- FastAPI
- DFCCNet
- Density regression
- IP cameras
- SMS alerts
- Company
FocalDive- Role
- AI Engineer
- Status
- In progress
Challenge
Birds overlap heavily, so counting them by detection is unreliable, and memory use grew with every camera added until inference ran out of memory.
Approach
Chose density regression over object detection to count overlapping birds accurately, and standardised frames to 600 px behind a single inference queue.
Results
Headcounts now run continuously with automated alerts, out-of-memory failures are gone, and memory stays flat as more cameras come online.