self-supervised learning

This work proposes the conditional slot initialization using clustering algorithms instead of random initialization and allows to generate flexible number of slots. Furthermore, it analyzes the effect of permutation symmetry including invariance and equivariance on the object-centric slot representations indicating the permutation equivariant mean-shift model presents notable advances especially for complex scenes.

In this work, we propose a few-shot pose estimation (FSPE) approach called SA6D, which uses a self-adaptive segmentation module to identify the novel target object and construct a point cloud model of the target object using only a small number of cluttered reference images.