Steel surface defect detection

The brief

Steel sheets are inspected by eye at high speed, and defects slip through. Train an image model that finds and outlines surface defects by type, so inspectors can focus on flagged sheets. Good looks like: segmentation quality per defect type, a study of missed defects, and an overlay demo.

What you will build

1. Explore the images and defect masks
2. Train a segmentation model (U-Net) with a pre-trained encoder
3. Evaluate with Dice score per defect class
4. Study missed and false defects
5. Build an overlay demo

Free, open data to build it with

• Severstal steel defect detection ↗ Kaggle

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.