Build a GAN-Based Defect Generator for a Hardware Manufacturing Line
Overview
What this challenge is about.
You receive around 60,000 good-unit images and around 380 defective-unit images across 4 defect classes. Train a class-conditional GAN (StyleGAN2-ADA or a smaller alternative for the budget) on the defective set, generate around 2,000 synthetic defects per class, and retrain the existing classifier with the augmented data. Compare precision/recall on a held-out test set. Document failure modes of the GAN with a small visual inspection by a quality-engineering reviewer (you).
The Brief
What you'll do, and what you'll demonstrate.
Use GAN-synthesized defect images to lift the classifier's precision while keeping recall above 0.98.
Earning criteria — what you'll demonstrate
- Train class-conditional GANs on small imbalanced datasets
- Use synthetic data to address class imbalance defensibly
- Evaluate classifier improvements with per-class precision/recall
- Identify and document generative-model failure modes
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
Machine Learning Engineer
Shipping a GAN-augmented classifier improvement on a real manufacturing line is exactly the kind of high-stakes MLE work hardware companies hire for.
This challenge sharpens
- gans
- data-augmentation
- imbalanced-classification
Computer Vision Engineer
Class-conditional GANs on imbalanced visual defect data are common in CV engineer work at hardware manufacturers.
This challenge sharpens
- gans
- class-conditional-generation
- evaluation
ML Researcher
Documenting GAN failure modes with quality-engineering review is the kind of empirical honesty hiring committees look for in ML research.
This challenge sharpens
- gans
- class-conditional-generation
- evaluation