AutoLay

AutoLay是由机器人研究中心和IIIT Hyderabad共同创建的数据集，专注于自动驾驶中的模态布局估计任务。该数据集包含超过16000张图像，覆盖12公里的行驶距离，涵盖多种城市环境。数据集内容包括人行道、车辆、斑马线和车道的高质量3D标注，以及鸟瞰图和图像空间中的标注。创建过程中，利用激光雷达扫描和精确的里程计信息进行全局视图注册，半自动化数据标注流程以减少人工工作量。AutoLay的应用领域主要集中在自动驾驶的感知问题上，旨在通过模态场景布局估计提高对遮挡和截断实体的理解，从而推动自动驾驶技术的发展。

AutoLay is a dataset co-developed by the Robotics Research Center and IIIT Hyderabad, focusing on the task of modal layout estimation in autonomous driving. This dataset contains over 16,000 images, spans a driving distance of 12 kilometers, and covers diverse urban environments. It includes high-quality 3D annotations for sidewalks, vehicles, zebra crossings, and lanes, as well as annotations in both bird's-eye view and image space. During its creation, LiDAR scans and precise odometry information were utilized for global view registration, and a semi-automated data annotation pipeline was adopted to reduce manual workload. The application scope of AutoLay mainly centers on perception tasks in autonomous driving, aiming to enhance the understanding of occluded and truncated entities via modal scene layout estimation, thereby advancing the development of autonomous driving technologies.