MSP60K
收藏资源简介:
MSP60K是由安徽大学创建的一个大规模跨域行人属性识别数据集,包含60,122张图像和57个属性标签,涵盖八个不同场景。数据集通过智能监控系统和手机收集,经过模糊、遮挡、光照变化等破坏性操作处理,以更好地模拟真实世界挑战。数据集的创建旨在解决现有数据集性能接近饱和、忽视跨域影响等问题,适用于行人检测、跟踪等以人为中心的任务,推动行人属性识别模型的发展和实际部署。
MSP60K is a large-scale cross-domain pedestrian attribute recognition dataset created by Anhui University. It consists of 60,122 images and 57 attribute labels, spanning eight distinct scenarios. The dataset is collected via intelligent surveillance systems and mobile phones, and has been subjected to degradations such as blurring, occlusion, and illumination variations to better mimic real-world challenges. It was developed to resolve the shortcomings of existing datasets, namely their nearly saturated performance and the oversight of cross-domain impacts. This dataset is applicable to human-centric tasks including pedestrian detection and tracking, and contributes to advancing the development and real-world deployment of pedestrian attribute recognition models.

- 1Pedestrian Attribute Recognition: A New Benchmark Dataset and A Large Language Model Augmented Framework安徽大学 · 2024年



