Image-Classification Model for a Quality-Control Line at a Bottling Plant
Overview
What this challenge is about.
Train an image classifier on 8,000 labeled bottle images (3 defect classes + 'ok'). Use transfer learning from a pre-trained backbone (EfficientNet-B0 or MobileNetV3) — the line's Jetson Nano cannot run a large model. Compare against a baseline (logistic regression on hand-engineered features) and a from-scratch small CNN. Evaluate: per-class precision/recall, confusion matrix, latency on Jetson Nano, false-positive cost vs false-negative cost. Recommend a deployment architecture (camera placement, on-device inference vs edge box, fallback to human on low-confidence). Deliver: training code + saved model, evaluation report (6 pages), Jetson Nano latency benchmark, 5-page deployment recommendation.
The Brief
What you'll do, and what you'll demonstrate.
Build a deep-learning bottle-defect classifier that hits 95 percent recall on each defect class at under 200ms latency on Jetson Nano.
Earning criteria — what you'll demonstrate
- Apply transfer learning to a constrained-hardware computer-vision task
- Compare deep-learning vs feature-engineered baselines honestly
- Evaluate models with cost-asymmetric metrics (false positives vs false negatives)
- Recommend a deployment architecture respecting edge-hardware limits
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.
Career mappings coming soon.