Occluder image dataset

This dataset contains a range of RGBA occluder masks and photographs that can be easily superimposed onto other images or fed into a dataloader for training computer vision models. Each directory contains a different occluder type, with subdirectories indicating the approximate percentage (+/- 5%) of the image space covered by the occluders. For instance, occluders in a directory named '40' will leave an underlying image ~60% visible when superimposed over it. Please cite the following conference abstract if you use this dataset for publication (manuscript reference coming late 2024):

David Coggan, Frank Tong; Unsupervised contrastive learning and supervised classification training have opposite effects on the human-likeness of CNNs during occluded object processing. Journal of Vision 2023; 23(9):5448. https://doi.org/10.1167/jov.23.9.5448

Download