arXiv 2603.097602026-07-17cs.CVcs.ROeess.IV

PanoAffordanceNet：迈向360°室内环境中的整体 affordance 地标

PanoAffordanceNet: Towards Holistic Affordance Grounding in 360° Indoor Environments

School of Artificial Intelligence and Robotics, Hunan University, China（湖南大学人工智能与机器人学院）

Guoliang Zhu, Wanjun Jia, Caoyang Shao, Yuheng Zhang, Zhiyong Li, Kailun Yang

备注：: Accepted to IEEE/RSJ IROS 2026. The source code and benchmark dataset will be made publicly available at https://github.com/GL-ZHU925/PanoAffordanceNet

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

PanoAffordanceNet通过DASM和OSDH解决360°室内环境中的整体affordance地标问题，整合多级约束抑制语义漂移，构建360-AGD数据集，提升具身智能的场景感知能力。

摘要：

在全球感知对于具身智能体在360°空间中至关重要，但当前的 affordance 地标仍主要以物体为中心，且局限于视角视图。为弥合这一差距，我们引入了一个新的任务：360°室内环境中的整体 affordance 地标。该任务面临独特的挑战，包括来自等距投影（ERP）的严重几何失真、语义分散和跨尺度对齐困难。我们提出了PanoAffordanceNet，一个端到端的框架，其特征是用于纬度依赖校准的Distortion-Aware Spectral Modulator（DASM）和用于从稀疏激活中恢复拓扑连续性的Omni-Spherical Densification Head（OSDH）。通过整合多级约束，包括像素级、分布级和区域文本对比目标，我们的框架在低监督条件下有效抑制语义漂移。此外，我们构建了360-AGD，第一个高质量的全景affordance地标数据集。大量实验表明，PanoAffordanceNet显著优于现有方法，为具身智能中的场景级感知建立了坚实的基础。源代码和基准数据集将在https://github.com/GL-ZHU925/PanoAffordanceNet上公开。

英文摘要：

Global perception is essential for embodied agents in 360° spaces, yet current affordance grounding remains largely object-centric and restricted to perspective views. To bridge this gap, we introduce a novel task: Holistic Affordance Grounding in 360° Indoor Environments. This task faces unique challenges, including severe geometric distortions from Equirectangular Projection (ERP), semantic dispersion, and cross-scale alignment difficulties. We propose PanoAffordanceNet, an end-to-end framework featuring a Distortion-Aware Spectral Modulator (DASM) for latitude-dependent calibration and an Omni-Spherical Densification Head (OSDH) to restore topological continuity from sparse activations. By integrating multi-level constraints comprising pixel-wise, distributional, and region-text contrastive objectives, our framework effectively suppresses semantic drift under low supervision. Furthermore, we construct 360-AGD, the first high-quality panoramic affordance grounding dataset. Extensive experiments demonstrate that PanoAffordanceNet significantly outperforms existing methods, establishing a solid baseline for scene-level perception in embodied intelligence. The source code and benchmark dataset will be made publicly available at https://github.com/GL-ZHU925/PanoAffordanceNet.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。