The Composed Caltech dataset is a synthetic dataset generated from Caltech-256 images. It consists of images formed by "pasting" scaled, translated, and rotated versions of the original Caltech-256 images onto other images from the dataset. We divide the images from CALTECH into three multi- category object datasets -- CALTECH-2, CALTECH-3 and CALTECH-4 -- each consisting of 10,000 images. They have been created by concatenating two, three, and four objects, randomly selected from 20 predefined candidate categories, respectively.