CODA：用于自动驾驶目标检测的真实道路拐角案例数据集

CODA is the world's first real-world self-driving corner case dataset of 1500 scenes (frames) containing nearly 6K corner cases. CODA is split into a validation set of 1000 images and a test set of 500 images. The validation set contains 4008 objects of 27 object categories, whereas the test set contains 1929 objects of 34 object categories, including 7 categories absent in the test set. Corner case annotations are stored in "val/corner_case.json" in COCO-compatible format. Out of the 1000 scenes of the validation set, 717 are taken from ONCE, 89 are taken from nuScenes, and 194 are taken from KITTI. Due to license issues, for nuScenes and KITTI, only corner case annotations and the correponding sample indices/tokens of original datasets are provided ("val/kitti_indices.json" and "val/nuscenes_sample_tokens.json"). For ONCE, in addition to corner case annotations, we also provide the front-view images captured by the camera named "cam03". The images taken from onCE are named in the format of "[sequence_id]_[frame_id].jpg" (000001_1616005007200.jpg, for example). The two identifiers ("sequence_id" and "frame_id") can be used to extract other data (e.g., lidar point clouds) from the onCE dataset if needed. The annotation file keeps consistent with the COCO format and contains three keys: "images", "categories" and "annotations". Image domain tags (i.e., periods and weather conditions) and 2D bounding boxes with classes for all CODA images. CODA annotation can be grouped into 7 super-categories including pedestrian, cyclist, vehicle, animal, traffic facility, obstruction and misc, which can be further divided into 34 fine-grained categories. Moreover, these categories can also be divided into two collections, namely 1) instances of novel classes and 2) novel instances of common classes. As the names suggest, common classes stand for common object categories annotated by existing autonomous driving benchmarks, such as cars and pedestrians, whereas novel classes stand for the opposites, such as dogs and strollers. CODA also provides domain tags for all images including the periods and weather conditions. Specifically, we annotate the period tags to be either day or night and select the weather condition tags from sunny, cloudy and rainy. We hope the image domain tags can help researchers dig into the underlying reasons of corner cases for reliable object detection.

CODA数据集是全球首个真实世界自动驾驶边缘场景（Corner Case）数据集，包含1500个场景（帧），近6000个边缘场景样本。该数据集被划分为包含1000张图像的验证集，以及包含500张图像的测试集。验证集涵盖27个物体类别，共4008个标注物体；测试集涵盖34个物体类别，共1929个标注物体，其中包含7个未在验证集中出现的类别。边缘场景标注以COCO兼容格式（COCO-compatible format）存储于"val/corner_case.json"中。在验证集的1000个场景中，717个来自ONCE数据集，89个来自nuScenes数据集，194个来自KITTI数据集。由于授权协议限制，针对nuScenes与KITTI数据集，仅提供边缘场景标注以及原始数据集对应的样本索引与Token（标记符），存储于"val/kitti_indices.json"与"val/nuscenes_sample_tokens.json"中。对于ONCE数据集，除边缘场景标注外，还提供名为"cam03"的摄像头采集的前视图图像。来自ONCE的图像命名格式为"[sequence_id]_[frame_id].jpg"（例如000001_1616005007200.jpg）。若有需要，可通过这两个标识符（sequence_id序列ID与frame_id帧ID）从ONCE数据集中提取其他数据（如激光雷达点云）。该标注文件与COCO格式保持一致，包含"images"、"categories"与"annotations"三个键。所有CODA图像均配有图像域标签（即时段与天气状况）以及带类别信息的二维边界框。CODA的标注可划分为7个超类别，包括行人（pedestrian）、骑行者（cyclist）、车辆（vehicle）、动物（animal）、交通设施（traffic facility）、障碍物（obstruction）及其他（misc），并可进一步细分为34个细粒度类别。此外，这些类别还可划分为两类集合：1）新类别样本；2）常见类别的新样本。顾名思义，常见类别指现有自动驾驶基准数据集已标注的常见物体类别，例如汽车与行人；而新类别则与之相反，例如犬类与婴儿车。CODA还为所有图像提供了时段与天气状况两类域标签。具体而言，我们将时段标签标注为白天或夜间，天气状况标签则从晴天、多云与雨天中选取。我们期望这些图像域标签能够帮助研究者深入探究边缘场景背后的成因，以实现更可靠的目标检测。

搜集汇总

背景与挑战

背景概述

CODA是一个专注于自动驾驶目标检测的真实道路拐角案例数据集，包含1500个场景和近6000个案例，分为验证集和测试集。数据集提供详细的2D边界框标注、34个细粒度类别及域标签（如时间段和天气条件），标注格式兼容COCO。