G-FARS: Gradient-Field-based Auto-Regressive Sampling for 3D Part Grouping

This paper proposes a novel task named "3D part grouping". Suppose there is amixed set containing scattered parts from various shapes. This task requiresalgorithms to find out every possible combination among all the parts. Toaddress this challenge, we propose the so called Gradient Field-basedAuto-Regressive Sampling framework (G-FARS) tailored specifically for the 3Dpart grouping task. In our framework, we design a gradient-field-basedselection graph neural network (GNN) to learn the gradients of a logconditional probability density in terms of part selection, where the conditionis the given mixed part set. This innovative approach, implemented through thegradient-field-based selection GNN, effectively captures complex relationshipsamong all the parts in the input. Upon completion of the training process, ourframework becomes capable of autonomously grouping 3D parts by iterativelyselecting them from the mixed part set, leveraging the knowledge acquired bythe trained gradient-field-based selection GNN. Our code is available at:https://github.com/J-F-Cheng/G-FARS-3DPartGrouping.