LGAfford-Net: A Local Geometry Aware Affordance Detection Network for 3D Point Clouds

Ramesh Ashok Tabib, Dikshit Hegde, Uma Mudenagudi; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2024, pp. 5261-5270

In this paper we introduce LGAfford-Net a novel architecture tailored for affordance detection in 3D point clouds. Affordance crucial for human-robot interaction denotes regions on objects where interaction is possible. Understanding affordance demands perceiving 3D space akin to humans. Leveraging the ubiquity of point clouds in capturing 3D environments our method addresses challenges posed by their sparse unordered and unstructured nature. Unlike prior approaches that overlook local context and semantic cues we propose a Semantic Geometric Correlator (SGC) block integrating Local Geometric Descriptor (LGD) for local understanding and Edge Convolution for semantic awareness. The integration of SGC LGD and edge convolution within our network enhances its capability to perceive and understand affordances by leveraging both geometric and semantic information effectively. Additionally we employ Class Specific Classifiers (CSC) to accommodate multiple affordance types per point. CSC effectively establish one to many relationship between point to affordance labels. We demonstrate the results of proposed architecture on 3DAffordanceNet a benchmark dataset and compare them with state-of-the-art methods. We demonstrate the effectiveness of the features learnt by our proposed architecture for the point cloud classification task using the ModelNet40 dataset.