ShapeClipQuant: A Synthetic Dataset of Partially Clipped Geometric Objects for Visible-Fraction and Shape-Completeness Estimation

## Purpose ShapeClipQuant is a controlled synthetic dataset of partially clipped filled geometric objects for visible-fraction estimation, shape-completeness analysis, classification, regression, and controlled robustness diagnostics. ## Composition The four classes are triangle, rectangle, square, and octagon. The release contains 25,600 grayscale 64×64 PNG images, 25,600 binary masks, and 25,600 metadata rows: train 16,000, validation 3,200, and test 6,400. ## Image and clipping properties Images are 64×64 grayscale PNGs with filled shapes. Clipping levels are complete (0.99–1.00), mild (0.85–<0.99), moderate (0.70–<0.85), and severe (0.60–<0.70). `visible_ratio` is the analytical geometric ratio of nominal-polygon area inside the image frame, rather than a raster foreground fraction.