Vehicle Damage Detection
Per-panel damage segmented and classified from a photo of the vehicle.
- Client
- Confidential
- Year
- 2024
- Category
- Computer Vision
- Role
- Design + Build
- Timeline
- 4–6 weeks
The problem
Damage assessment from photos is slow and inconsistent: two assessors, two answers, and a growing backlog.
What we built
A Faster R-CNN instance-segmentation model built on Meta’s Detectron2 outlines each damaged region and classifies it (dent, scratch, break). A Streamlit interface lets an assessor upload a photo and get the segmented result with confidence scores for review.
The result
A consistent first pass on every photo, so assessors spend their time on the edge cases instead of the obvious ones.
- pixel mask + class for each damaged area
- per regionpixel mask + class for each damaged area
- no fixed rig, works from a phone photo
- one photono fixed rig, works from a phone photo
- assessor confirms or corrects, fast
- review UIassessor confirms or corrects, fast
Built with
- Python
- Detectron2
- Faster R-CNN
- OpenCV
- Streamlit
We deliver what we commit.
Tell us what you're trying to build.
We'll come back within 24 hours with honest feedback on scope, timeline and cost, whether or not we turn out to be the right fit.