3x3 Pixel Pattern Classification Dataset: 12 Adjacent Pair Classes

This synthetic dataset is designed for pattern recognition tasks using 3×3 pixel RGB images. It contains 12 distinct classes, each representing a unique adjacent pixel pair highlighted in dark yellow (Goldenrod) against a background of random dark colours. The dataset is ideal for benchmarking classical CNNs, Quantum CNNs (QCNNs), and hybrid quantum-classical models due to its minimal size and clear structural patterns. 1,200 images, 12 classes, 3x3 pixels. Version v2: All non-yellow pixels use 7 unique dark colors per image.