ClevrTex

We learn object-centric representations in an unsupervised manner by differentiably parameterizing and solving a min-cut problem to partition the image. Published at ICLR 2023.