Vehicle damage assessment from photos

The brief

Motor claims wait days for a surveyor to see minor dents. Train an image model that finds the damaged area in a customer's photo, names the part and rates the damage as minor, moderate or severe, so simple claims can be fast-tracked. Good looks like: detection accuracy per damage type, a severity rating checked against labels, and a demo for uploaded photos.

What you will build

1. Prepare the labelled damage images
2. Train a detection or segmentation model for damage
3. Add a severity classifier
4. Evaluate per damage type and severity
5. Build an upload-and-assess demo with confidence scores

Free, open data to build it with

• CarDD: car damage detection dataset ↗ University of Science and Technology of China

Check each dataset's licence before you publish your work. Your mentor gives your team its own version of the brief (data, scope or region), so every team's project is its own.