Computer-vision models detect and classify pavement defects and pre-label the images. Engineers review, correct and approve the labels, which then become training data.
Computer vision
Human-in-the-loop
React
TypeScript
Use case
Road-inspection imagery has to be labeled for cracks, potholes and surface distress before it’s useful, and labeling it all by hand is slow.
Constraints
Engineers stay in control: every model label is reviewed by a person.
Label quality has to stay consistent enough to train on.
What I built
Integration of the computer-vision models that detect and classify defects.
Automatic pre-labels generated from inspection imagery.
A human-in-the-loop workflow where engineers review, correct and approve model labels.
QA review and dataset-management flows that organize labeled data for training.
The React and TypeScript application around it.
Engineers correct the model’s first pass instead of labeling every image from scratch.
