nuInsSeg

该数据集名为nuInsSeg，是首个集成了激光雷达点云、RGB图像、3D边界框、2D边界框以及与2D和3D边界框注释一致的手动标注实例掩膜的数据集。该数据集涵盖了10个类别：汽车、卡车、公交车、拖车、施工车辆、行人、摩托车、自行车、交通锥和障碍物。在规模上，训练集包含了168,780张图像，带有789,193个实例掩膜标注；验证集则包含36,114张图像，带有157,879个掩膜标注。该数据集的任务是实例分割。

The dataset named nuInsSeg is the first of its kind to integrate LiDAR point clouds, RGB images, 3D bounding boxes, 2D bounding boxes, and manually annotated instance masks that are aligned with the corresponding 2D and 3D bounding box annotations. It covers 10 categories: car, truck, bus, trailer, construction vehicle, pedestrian, motorcycle, bicycle, traffic cone, and obstacle. In terms of scale, the training set contains 168,780 images with 789,193 instance mask annotations, while the validation set includes 36,114 images with 157,879 mask annotations. The task of this dataset is instance segmentation.

搜集汇总

背景与挑战

背景概述

nuInsSeg是首个集成激光雷达点云、RGB图像、3D与2D边界框及一致手动标注实例掩膜的数据集，涵盖汽车、行人等10个类别；训练集包含168,780张图像和789,193个实例掩膜，验证集有36,114张图像和157,879个掩膜，用于实例分割任务，规模较大且标注丰富。