Case study · Retail
Two million shelf images for a North American grocer
Product-level detection on shelf photographs from 900 stores, so that empty shelves and misplaced stock are seen in minutes rather than days.
- Shelf images labelled
- 2.1 million
- Distinct products
- 48,000
- Turnaround
- Overnight
The work
Store staff photograph each aisle twice a day. Our labellers draw a box around every product facing and match it to the chain’s catalogue of 48,000 products, including the many that differ only in flavour or size.
Work runs in both centres so that images taken during the North American afternoon are labelled overnight. Manila and Nairobi teams calibrate against the same gold set every week.