Machine Learning
Casting Surface Defect Detection with Grad-CAM
A defect-detection project centered on model visibility, not just classification output.
Computer VisionGrad-CAMCNN
Overview
Focused on surface defect detection in a computer-vision setting.
Used Grad-CAM to make model attention easier to inspect and communicate.
Treated interpretability as part of the product value of the system, not just a research extra.
Results
Defect classifier with Grad-CAM heatmaps for visual interpretability
Trained on industrial casting surface dataset
Heatmap overlays make model decisions inspectable by non-ML stakeholders
Key Outcomes
Applied Grad-CAM to surface model reasoning more clearly.
Worked on vision classification in an industrial inspection context.
Balanced technical modeling with explainability needs.
Need the full implementation story?
Walkthrough, private repo access, or implementation details available.