CAR (Cityscapes Attributes Recognition)

CAR 包含 Cityscapes 数据集中对象的视觉属性。对于图像中的每个对象，我们都有一个依赖于对象类别的属性列表。例如，车辆类别具有可见性属性，而行人具有活动属性（步行、站立等）。该数据集的目的是简化自动驾驶车辆更好算法的开发，因为这需要全面了解整个场景及其所有细节，包括所有对象的属性。我们选择 Cityscapes，因为它已经包含不同类型的有用注释，并且添加属性将消除开发具有自我或半监督的算法的巨大负担。

The CAR dataset holds visual attributes for objects within the Cityscapes dataset. For every object present in an image, we offer a category-specific attribute list. For instance, the vehicle category includes the visibility attribute, while the pedestrian category encompasses activity attributes such as walking, standing, and similar states. The core objective of this dataset is to facilitate the development of enhanced algorithms for autonomous vehicles, which necessitate a comprehensive grasp of the full scene and all its intricate details, including the attributes of every object within it. We chose Cityscapes as the foundational dataset due to its already existing diverse set of valuable annotations, and adding these attributes will remove the considerable burden of developing self-supervised or semi-supervised algorithms.

搜集汇总

背景与挑战

背景概述

CAR数据集基于Cityscapes数据集，为图像中的对象提供视觉属性，例如车辆的可见性和行人的活动状态。该数据集旨在支持自动驾驶算法的开发，通过补充现有注释来减轻算法设计的负担。