KITTI INStance dataset (KINS)

We annotate a total of 14, 991 images from KITTI to form a large-scale amodal instance dataset, namely KINS. The dataset is split into two parts where 7, 474 images are for training and the other 7, 517 are for testing. All images are densely annotated with instances by three skilled annotators. The annotation includes amodal instance masks, semantic labels and relative occlusion order, from which inmodal instance masks can be easily inferred.