Learns disentangled representations of scenes by decomposing them into object-centric concepts in an unsupervised way.
