Synthetic Semiconductor Wafer Micro-Defect Detection Dataset

This dataset contains synthetic grayscale images of semiconductor wafers with periodic circuit patterns, designed for unsupervised anomaly detection benchmarking in computer vision. The training set contains 200 normal (non-defective) wafers. The test set contains 100 wafers (50 normal, 50 defective). Defects are subtle phase perturbations in the periodic pattern that are invisible in the raw spatial domain and require frequency-domain analysis (2D FFT) to detect. Generation method: Python 3.11 with NumPy and PIL. Fully reproducible with fixed random seed (42). Files: - train/ : 200 normal images (256×256 PNG) - test/ : 100 images (50 normal, 50 defective) - train.csv : Training labels - test_answers.csv : Ground truth - generate_dataset.py : Reproducible generation script License: CC BY 4.0 (Commercial use allowed)