CARLA-BSP

CARLA-BSP数据集是由西班牙巴塞罗那的计算机视觉中心(CVC)创建的，专注于模拟环境中的行人数据。该数据集包含近400个场景，每个场景记录了900帧的视频数据，涵盖了单个行人的行为，如穿越街道等。数据集不仅包括RGB视频，还提供了语义分割数据和详细的2D及3D姿态关键点，适用于行人检测、姿态估计和行为分析等研究。通过CARLA模拟器生成，CARLA-BSP数据集为研究者提供了一个高质量、可控的实验平台，以推动自动驾驶和计算机视觉技术的发展。The CARLA-BSP dataset was developed by the Computer Vision Center (CVC) in Barcelona, Spain, and focuses on pedestrian data in simulated environments. This dataset contains nearly 400 scenarios, with each scenario recording 900 frames of video data covering the behaviors of individual pedestrians such as crossing streets. The dataset not only includes RGB videos, but also provides semantic segmentation data and detailed 2D and 3D pose keypoints, which is applicable to research areas including pedestrian detection, pose estimation and behavior analysis. Generated via the CARLA simulator, the CARLA-BSP dataset provides researchers with a high-quality and controllable experimental platform to advance the development of autonomous driving and computer vision technologies.

搜集汇总

背景与挑战

背景概述

CARLA-BSP是一个用于自动驾驶研究的模拟数据集，专注于行人过街行为。它包含近400个视频，每个视频900帧，分辨率为1600×600，标注了行人是否正在过街，并提供了详细的元数据，如行人姿态、年龄、性别等。数据集由ARCANE项目创建，但包含一些未清理的渲染问题，如缺失帧和颜色错误。