SynFog
收藏资源简介:
SynFog数据集是由清华大学和斯坦福大学合作创建的,旨在通过端到端成像模拟生成逼真的雾天图像,以推动自动驾驶中的实际去雾技术。该数据集包含500个独特的户外场景,每个场景在自然天空光和主动光源(如街灯和汽车照明)下拍摄,涵盖三种不同浓度的雾,对应可见度分别为600米、300米和150米。此外,每个场景还提供了像素级精确的深度数据和分割标签。数据集的创建过程涉及使用体积路径追踪进行雾场景渲染,并通过物理基础的相机模型处理场景辐射数据,以忠实再现真实相机设备。SynFog数据集主要应用于解决自动驾驶中的视觉感知和检测精度问题,特别是在雾天条件下的图像去雾任务。
SynFog Dataset was collaboratively developed by Tsinghua University and Stanford University, with the goal of generating photorealistic foggy images through end-to-end imaging simulation to advance practical dehazing technologies in autonomous driving. This dataset contains 500 unique outdoor scenes, each captured under both natural skylight and active light sources such as street lamps and vehicle lights, covering three distinct fog concentrations corresponding to visibility levels of 600 meters, 300 meters, and 150 meters respectively. In addition, pixel-accurate depth data and segmentation labels are provided for every scene. The creation of the dataset involves rendering foggy scenes using volumetric path tracing, and processing scene radiance data with physics-based camera models to faithfully reproduce real-world camera devices. The SynFog dataset is primarily applied to address challenges related to visual perception and detection accuracy in autonomous driving, particularly the image dehazing task under foggy conditions.




