HM3D-OVON
收藏资源简介:
HM3D-OVON数据集是由佐治亚理工学院创建的,旨在扩展对象目标导航(ObjectNav)任务的语义范围。该数据集包含超过15,000个家庭对象的注释实例,涵盖379个不同的类别,这些数据来源于真实世界的3D扫描环境。与传统的ObjectNav数据集不同,HM3D-OVON支持在测试时通过自由形式的语言定义开放词汇的目标对象。数据集的创建过程包括对真实世界环境的3D扫描和对象注释,确保了数据的真实性和多样性。HM3D-OVON主要应用于开发能够根据自由形式语言指令在真实世界环境中导航并找到指定对象的机器人助手,旨在解决对象导航任务中的开放词汇问题。
The HM3D-OVON dataset was created by the Georgia Institute of Technology to expand the semantic scope of the object goal navigation (ObjectNav) task. It contains over 15,000 annotated instances of household objects spanning 379 distinct categories, with all data sourced from real-world 3D scanned environments. Unlike traditional ObjectNav datasets, HM3D-OVON supports defining open-vocabulary target objects via free-form natural language during testing. The dataset's creation process includes 3D scanning of real-world environments and object annotation, ensuring the authenticity and diversity of the collected data. HM3D-OVON is primarily applied to develop robotic assistants that can navigate real-world environments and locate specified objects based on free-form natural language instructions, aiming to address the open-vocabulary challenge in the object navigation task.

- 1HM3D-OVON: A Dataset and Benchmark for Open-Vocabulary Object Goal Navigation佐治亚理工学院 · 2024年



