Checklist NOAI Singapore 2026 Final Assessment · Task 2
The Segmentation Fracture (Debugging U-Net)
Repair skip-connection shape mismatches in a padding-free U-Net and re-implement mean IoU from scratch.
The task
Programming Task 2 (Deep Learning – Computer Vision, 20 marks) describes a U-Net that segments satellite images into Urban (0) and Forest (1). An intern removed padding, so encoder feature maps no longer match decoder feature maps.
Part 1 (10 marks) implements center_crop so that skip connections can be concatenated and asks for a short discussion of the trade-offs of cropping. Part 2 (10 marks) re-implements mean Intersection over Union without scikit-learn. Part 3 is an unmarked integrity check.
Abridged by SOTA from the official materials. The official statement has the exact rules, and it wins wherever this summary differs.
At a glance
- You get
- A mock random tensor of shape (B, 3, 256, 256).
- You submit
- The completed notebook.
- Scoring
- Marks per part as stated in the notebook.
- Rules
- Allowed: torch, torch.nn, torch.nn.functional, numpy.
- Banned: sklearn, torchmetrics and other high-level metric wrappers.
- Format
- NOAI 2026 Final Assessment, 6 March 2026, 9:30 am–12:30 pm (3 hours, 100 marks): Section 1 (20 MCQs, 20 marks) and three programming tasks. Individual.