ClevrTex Unsupervised Multi-object Segmentation Dataset

ClevrTex is a synthetic image dataset for comparing, evaluating, and analyzing unsupervised multi-object segmentation algorithms. The dataset contains 5,0000 images created from 60 materials, depicting 3-10 objects arranged on a synthetic background. The test set has 1,0000 images created from 25 different materials. The synthetic scenes in this dataset are obtained based on physically relevant techniques, with different shapes, textures, and photo-mapped materials.