VitralColor-12: A Synthetic Twelve-Color Segmentation Dataset from GPT-Generated Stained-Glass Images

VitralColor-12, a synthetic dataset for color classification and segmentation, utilizes LLMs in specific GPT-5 and DALL·E 3 models to generate images of stained-glass. This approach simplifies labeling by using the dark steel structure supporting the glass as a guide, which provides easy regions to label with a single color per region. After that, we obtain the images and use at least one hand-labelled centroid per color to automatically cluster all pixels based on Euclidean distance and morphological operations, including erosion and dilation with two iterations per operation, and a kernel size of 2 pixels. This process enables us to label a classification dataset and generate segmentation maps automatically. Our dataset comprises 910 images, organized into 70 generated images and 12 pixel segmentation maps—one for each color, which includes 9,509,524 labeled pixels, 1,794,758 of which are unique. We include tables with pixel values in RGB, HSL, CIELAB, and YCbCr color representations, enabling detailed color analysis and training of machine learning algorithms in different color spaces. Furthermore, we also included descriptive statistics and ΔE76, ΔE94, and CIELAB a vs b Chromacity, which prove the distribution, applicability, and realistic perceptual structures, including warm, neutral, and cold colors, as well as the high contrast between black and white colors, offering meaningful perceptual clusters, reinforcing its utility for color segmentation and classification.