Spatially Invariant Unsupervised 3D Object-Centric Learning and Scene Decomposition
This addresses the problem of scalable machine intelligence for high-level relational reasoning in 3D scenes, though it appears incremental as it builds on prior object-centric learning methods.
The paper tackles unsupervised object-centric learning on 3D point clouds by introducing SPAIR3D, a framework that factorizes scenes into spatial mixture models, resulting in scalable detection and segmentation of an unknown number of objects without supervision.
We tackle the problem of object-centric learning on point clouds, which is crucial for high-level relational reasoning and scalable machine intelligence. In particular, we introduce a framework, SPAIR3D, to factorize a 3D point cloud into a spatial mixture model where each component corresponds to one object. To model the spatial mixture model on point clouds, we derive the Chamfer Mixture Loss, which fits naturally into our variational training pipeline. Moreover, we adopt an object-specification scheme that describes each object's location relative to its local voxel grid cell. Such a scheme allows SPAIR3D to model scenes with an arbitrary number of objects. We evaluate our method on the task of unsupervised scene decomposition. Experimental results demonstrate that SPAIR3D has strong scalability and is capable of detecting and segmenting an unknown number of objects from a point cloud in an unsupervised manner.