RaidaR
收藏资源简介:
RaidaR是一个为自动驾驶研究提供支持的丰富注释图像数据集,专注于雨天街道场景。该数据集包含迄今为止最大数量的雨天街道场景图像(58,542张),其中5,000张提供语义分割,3,658张提供对象实例分割。RaidaR图像涵盖了广泛的现实雨天引起的伪影,包括雾、水滴和道路反射,有效增强现有街道场景数据集,以改善雨天天气下的数据驱动机器感知。数据集通过结合手动分割和自动化处理,采用半自动方案进行高效标注,显著减少了标注时间。RaidaR不仅提升了现有分割算法的准确性,还通过引入一种新颖的非配对图像到图像翻译算法,直接受益于其注释数据集,为自动驾驶在恶劣天气条件下的视觉感知任务提供了强有力的数据支持。
RaidaR is a richly annotated image dataset supporting autonomous driving research, focusing on rainy street scenes. This dataset contains the largest number of rainy street scene images to date, with a total of 58,542 images, among which 5,000 are provided with semantic segmentation annotations and 3,658 with object instance segmentation annotations. Images in RaidaR cover a wide range of real-world rainy-induced artifacts, including fog, water droplets, and road reflections, effectively augmenting existing street scene datasets to improve data-driven machine perception under rainy weather conditions. The dataset adopts a semi-automatic annotation scheme by combining manual segmentation and automated processing, which significantly reduces annotation time. RaidaR not only improves the accuracy of existing segmentation algorithms, but also directly benefits from its annotated dataset by introducing a novel unpaired image-to-image translation algorithm, providing strong data support for visual perception tasks of autonomous driving under harsh weather conditions.

- 1RaidaR: A Rich Annotated Image Dataset of Rainy Street Scenes西蒙弗雷泽大学 · 2021年



