Street-level waste imagery was annotated every week so a European smart city consortium could train and improve waste recognition AI.
The challenge
The client needed thousands of waste images annotated weekly without slowing model improvement cycles.
Visual clutter, lighting variation and ambiguous objects required consistent labeling rules and quality review.
The approach
SBL set up a high-volume annotation workflow with clear class definitions and review loops.
The team handled weekly image batches, corrected edge cases and delivered datasets for model retraining.
Classify
Define waste categories and labeling rules.
Annotate
Process weekly street-level image batches.
Improve
Use review feedback to stabilize future model data.
The result
The AI program gained a more economical and dependable path to keep waste recognition models supplied with training data.
- More than 5,000 images were handled weekly.
- The operating model reduced annotation cost by about 60%.
- Recognition accuracy reached 91% in the documented program.
- Weekly dataset delivery supported ongoing model improvement.