AI & computer vision
Amenity detection with AI
A computer vision proof of concept that spots pools, tennis courts and playgrounds in real estate imagery.
- Client
- Real estate
- Services
- POC as a service, Computer vision, Machine learning
The challenge
Checking real estate listings for amenities by hand is slow and inconsistent. The goal was to show that AI could detect amenities at scale, cutting manual inspection time and improving accuracy.
What we built
We built a custom object detection pipeline on Detectron2, the PyTorch-based computer vision framework, covering data preparation, training and evaluation.
We collected and labelled a large image dataset using manual and semi-automated techniques, then fine-tuned the model with transfer learning for amenities such as swimming pools, tennis courts and playgrounds.
Evaluated on a separate test set, the proof of concept showed promising accuracy and a clear path to production use.