JHU-CROWD++
收藏资源简介:
JHU-CROWD++是由约翰斯·霍普金斯大学创建的大规模无约束人群计数数据集,包含4372张图像和151万个标注。该数据集在多种环境和场景下收集,特别考虑了天气影响和光照变化,使其成为极具挑战性的数据集。数据集提供了图像级和头部级的丰富标注,包括点级标注、近似尺寸、模糊级别等。JHU-CROWD++适用于评估和比较最新的人群计数方法,旨在解决复杂环境中的人群计数问题。
JHU-CROWD++ is a large-scale unconstrained crowd counting dataset developed by Johns Hopkins University. It comprises 4,372 images and over 1.51 million annotations. Collected across diverse environments and scenarios, this dataset takes deliberate consideration of weather impacts and illumination variations, rendering it a highly challenging benchmark for crowd counting research. It provides rich annotations at both image-level and head-level, including point-level head annotations, approximate head sizes, blur levels, and other relevant attributes. JHU-CROWD++ is designed to evaluate and compare state-of-the-art crowd counting methods, with the goal of addressing crowd counting challenges in complex real-world environments.




