IntPhys 2019
收藏资源简介:
IntPhys 2019是由法国国家科学研究中心和巴黎笛卡尔大学创建的视觉物理理解评估基准。该数据集包含15000个可能事件的视频,总时长约21小时,用于训练和测试机器对物理世界的理解能力。数据集通过游戏引擎生成,确保了视频内容的高质量和多样性。IntPhys 2019旨在通过测试系统对物理原则的理解,如物体永久性、形状恒常性和时空连续性,来推动人工智能在复杂视觉任务上的表现。该数据集的应用领域包括自动驾驶、工作场所或家庭机器人等,旨在解决机器在现实世界中对物理现象的理解和适应问题。
IntPhys 2019 is a visual physics understanding evaluation benchmark developed by the French National Centre for Scientific Research (CNRS) and Paris Descartes University. This dataset comprises 15,000 videos of hypothetical physical events, with a total duration of approximately 21 hours, and is used for training and evaluating machines' ability to comprehend the physical world. Produced using game engines, the dataset ensures high quality and diversity of its video content. IntPhys 2019 aims to advance the performance of artificial intelligence in complex visual tasks by testing systems' understanding of core physical principles including object permanence, shape constancy, and spatiotemporal continuity. Application scenarios of this dataset cover autonomous driving, workplace or home robotics, and other fields, with the objective of addressing challenges in machine understanding and adaptation to physical phenomena in the real world.

- 1IntPhys: A Framework and Benchmark for Visual Intuitive Physics Reasoning法国国家科学研究中心/巴黎笛卡尔大学 · 2020年



