
Computer Vision Research Intern
A*STAR — Agency for Science, Technology and Research
- Built an end-to-end CNN-based defect detection pipeline for AOG identification on wafer SEM images.
- Curated and pixel-annotated 288 high-resolution SEM images using CVAT to construct the training set.
- Benchmarked classical intensity baseline, plain UNet, and ResNet34-UNet — improved IoU, Dice, and F1 under identical settings.
- Deployed the model as an application prototype reporting defect count, location, area, and batch-level statistical charts.












