Geometrical shapes datasets
Description
Labeled datasets of images of geometrical shapes. Can be used for training and conceptual learning of supervised and unsupervised models. Use pickle, zip to extract raw images and labels. bw_shapes_3_1800 grayscale, 64x64, 3 distinct types (concepts), 600 per type 1800 total Grayscale circles, variation in shade, size, background Grayscale triangles, variation shade, size, background Empty (grey) backgrounds, variation in shade. Labels: 0 - 2 (circle, triangle, background) col_shapes_9_7200 color, 64x64, 9 distinct types (concepts), 800 per type 7200 total: - Two color (red, blue) circles, grey background, variation in shade, size, background - R,b triangles, grey background, variation shade, size, background - R,b horizontal stripes wide (red), narrow (blue) variation shade, width (r > 0.5 size, b < 0.5 size), background - R,b vertical stripes narrow (red), wide (blue) variation shade, width (r < 0.5 size, b > 0.5 size), background - Empty (grey) backgrounds, variation in shade. Labels 0,1 (circle r, b) 2,3 (triangle r, b) 4,5 (horizontal r,b) 6,7 (vertical r,b) 8 (background) Use pickle, zip to extract raw images and labels.
Files
Steps to reproduce
Use convolutional autoencoder to produce high quality generation of input images and a low-dimensional geometric embedding