In this work, we present SP \(\mathbb {A}\) N (semantic perception aware network), a novel approach for affordance detection in 3D point clouds. Affordance detection on 3D objects is crucial for enabling effective interaction, as it allows the perception and understanding of plausible actions with the object. The intricate and unstructured nature of 3D point cloud data presents challenges to traditional detection methods due to its high dimensionality, sparse representation, and limited contextual information. Integrating semantic perception in affordance detection becomes crucial to recognize object categories and their associated affordances, fostering context-aware interactions and enhancing comprehension within 3D environments. Towards this, we propose semantic perception aware network (SP \(\mathbb {A}\) N) which includes semantic abstraction encoder to comprehensively extract semantic perception features, and point affordance classifier to establish the mapping between multiple affordances to a point under consideration. Semantic abstraction encoder captures both global and detailed information by integrating local geometry and semantic cues at different levels of abstraction using semantic geometric approximator module. At every level of abstraction, we propose local geometric correlator to capture local geometric information and Weighted EdgeConv to enable semantic awareness in semantic geometric approximator. We demonstrate the effectiveness of SP \(\mathbb {A}\) N through extensive experiments on the 3D Affordance dataset and compare results with state-of-the-art methods.
