KITTI-360-APS

我们扩展了KITTI 360数据集，该数据集具有带有amodal panoptic注释的语义和实例标签，并将其命名为KITTI-360-APS数据集。它由九个城市街道场景序列组成，并为分辨率1408x376像素的61,168图像提供注释。我们的数据集包括10个类。如果类具有无定形区域或在任何时间点都无法移动，则我们将该类定义为东西。道路，人行道，建筑物，墙壁，围栏，杆子，交通标志，植被，地形和天空是这些东西。此外，数据集由7个事物类别组成，即汽车，行人，骑自行车的人，两轮车，货车，卡车和其他车辆。请注意，我们将自行车和摩托车类合并为一个称为两轮车的单一类。

We extend the KITTI 360 dataset, which is furnished with semantic and instance labels accompanied by amodal panoptic annotations, and designate it as the KITTI-360-APS dataset. This dataset comprises nine urban street scene sequences, and provides annotations for 61,168 images with a resolution of 1408 × 376 pixels. Our dataset encompasses a total of 10 classes. We define classes with amorphous regions or that are immovable at any time point as stuff classes. The 10 stuff classes are road, sidewalk, building, wall, fence, pole, traffic sign, vegetation, terrain, and sky. Additionally, the dataset includes 7 thing categories, namely car, pedestrian, cyclist, two-wheeler, van, truck, and other vehicles. Note that we merge the bicycle and motorcycle classes into a single class termed two-wheeler.

搜集汇总

背景与挑战

背景概述

KITTI-360-APS数据集是KITTI 360数据集的扩展版本，增加了amodal panoptic注释，包含9个城市街道场景序列和61,168张图像，涵盖10个'东西'类和7个'事物'类。该数据集由弗莱堡大学于2022年发布，适用于语义和实例分割研究。