SODA-D

SODA-D is a large-scale dataset tailored for small object detection in driving scenario, which is built on top of MVD dataset and owned data, where the former is a dataset dedicated to pixel-level understanding of street scenes, and the latter is mainly captured by onboard cameras and mobile phones. With 24704 well-chosen and high-quality images of driving scenarios, SODA-D comprises 277596 instances of 9 categories with horizontal bounding boxes.

SODA-D是一款专为驾驶场景小目标检测打造的大规模数据集，其构建依托于MVD数据集与自有数据。其中，MVD数据集是一款专注于街景像素级理解的专用数据集，而自有数据主要由车载摄像头与移动设备采集所得。该数据集共包含24704张经精心筛选的高质量驾驶场景图像，涵盖9个类别总计277596个目标实例，所有实例均配有水平边界框（horizontal bounding box）。