PANDA

PANDA是第一个gigaPixel级别的以人为中心的视频数据集，用于大规模，长期和多对象的视觉分析。PANDA中的视频是由gigapixel相机捕获的，涵盖了具有宽视野 (〜1千米 ² 面积) 和高分辨率细节 (〜gigapixel级别/帧) 的真实世界大型场景。场景可以包含具有超过100 × 比例变化的4k头部计数。PANDA提供丰富和分层的地面真相注释，包括15,974.6k边界框，111.8k细粒度属性标签，12.7k轨迹，2.2k组和2.9k交互。PANDA is the first gigaPixel-level human-centric video dataset designed for large-scale, long-term, and multi-object visual analysis. The videos in PANDA are captured by gigapixel cameras, covering real-world large-scale scenes with a wide field of view (~1 km² in area) and high-resolution details at ~gigaPixel per frame. The scenes may contain head counts with scale variations exceeding 100× and 4K-resolution details. PANDA provides rich and hierarchical ground-truth annotations, including 15,974.6k bounding boxes, 111.8k fine-grained attribute labels, 12.7k trajectories, 2.2k object groups, and 2.9k interaction instances.

搜集汇总

背景与挑战

背景概述

PANDA是一个gigaPixel级别的以人为中心的视频数据集，使用gigapixel相机捕获，覆盖宽视野和高分辨率细节的真实世界大型场景。该数据集提供了丰富的分层标注，包括边界框、属性标签、轨迹、组和交互，由清华大学和杜克大学于2020年发布。