Content Moderation
Boosting ML Efficiency for a Leading Insurance Provider
The Challenge
A leading European insurance company aimed to train a machine learning model to detect variations in rooftops—from dense urban clusters to isolated homes. The goal: improve the algorithm’s ability to recognize roof shapes, structures, and the presence of solar panels. The project required annotating over 30,000 satellite images from seven countries within six weeks. Achieving high accuracy despite image inconsistencies and varying quality demanded a scalable, efficient solution.
• • • •The Solution• • • •
DataForce addressed the challenge by combining our proprietary platform with innovative, scalable workflows:
- Training & Onboarding
- Annotators received live demos, detailed guidelines, and passed a qualification test to ensure accuracy before production.
- Customer Calibration
- A subset of images was annotated with full QA, reviewed with the client, and adjusted based on feedback—refining techniques for complex structures like high-density apartments.
- Quality Assurance Tracking
- Initial annotations underwent 100% QA, with oversight reduced as contributors demonstrated sustained accuracy, ensuring quality while improving speed.
- ML-Based Workflow Optimization
- A computer vision model trained on approved annotations generated automated pre-labels for annotators to refine, boosting efficiency without sacrificing precision.
Results
By combining human expertise with machine learning, DataForce delivered:
- 205,000+ individually annotated objects.
- 98.87% approval rate.
- Scalable, efficient annotation workflows.
Our client met their goals on time, processing a large volume of satellite imagery with high accuracy. We look forward to supporting future initiatives.