CODA

该数据集包含了大约1万个精心挑选的真实世界道路驾驶场景，这些场景中对29个代表性物体类别进行了高质边界框标注，形成了一个新颖的对象级边缘案例数据集。该数据集可与SODA10M结合使用，用于训练和验证，具有区分常见类别和新型类别的处理方式。规模上，该数据集大约包含10,000张图像，其任务专注于物体检测。

This dataset contains approximately 10,000 carefully curated real-world road driving scenarios, where 29 representative object categories are annotated with high-quality bounding boxes, forming a novel object-level edge case dataset. It can be used in conjunction with SODA10M for training and validation, and supports distinguishing between common and novel object categories. Comprising roughly 10,000 images, this dataset focuses on the task of object detection.