Train a GAN for Synthetic Defect Augmentation on a Factory Line
Overview
What this challenge is about.
You receive a labeled defect dataset (12 defect types, ranging from 8 to 4,200 examples each), the production classifier, and a starter StyleGAN2-ADA codebase. Train a GAN per rare class (those with fewer than 100 examples). Use the generated samples to augment the rare-class training data and retrain the classifier. Evaluate on a held-out real set with per-class recall and a confusion matrix. Report whether augmentation improves rare-class recall without hurting head-class precision. Write a 2-page memo for the line-quality team.
The Brief
What you'll do, and what you'll demonstrate.
Train per-class GANs to augment rare-defect data and quantify whether augmentation improves classifier rare-class recall on a body-panel line.
Earning criteria — what you'll demonstrate
- Train modern GANs on small-data regimes
- Use synthetic data to augment a real classifier and evaluate fairly
- Measure generation quality with FID + downstream task performance
- Communicate ML augmentation results to a non-AI quality team
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.
Computer Vision Engineer
Training GANs for industrial defect augmentation and proving downstream impact is exactly the day-one work of a CV engineer at any manufacturing-AI team.
This challenge sharpens
- gans
- stylegan
- image-generation
Machine Learning Engineer
Owning a synthetic-augmentation pipeline that ships measurable lift in a production classifier is core MLE work for industrial ML.
This challenge sharpens
- data-augmentation
- pytorch
- evaluation
Applied AI Scientist
Translating GAN augmentation results into a budget memo for a non-AI quality team is the applied-AI-scientist craft of bridging research and operations.
This challenge sharpens
- gans
- data-augmentation
- evaluation