This paper addresses the problem of semantic foggy scene understanding (SFSU). Due to the difficulty of collecting and annotating foggy images, the authors generate synthetic fog on real images depicting clear-weather outdoor scenes and use these partially synthetic data for SFSU with state-of-the-art convolutional neural networks (CNN). A complete pipeline is developed to add synthetic fog to real, clear-weather images using incomplete depth information. The authors apply their fog synthesis on the Cityscapes dataset and generate Foggy Cityscapes with 20550 images. SFSU is tackled in two ways: 1) with typical supervised learning, and 2) with a novel type of semi-supervised learning, which combines 1) with an unsupervised supervision transfer from clear-weather images to their synthetic foggy counterparts. Additionally, the usefulness of image dehazing for SFSU is studied. For evaluation, the authors present Foggy Driving, a dataset with 101 real-world images depicting foggy driving scenes, which come with ground truth annotations for semantic segmentation and object detection. Extensive experiments show that 1) supervised learning with synthetic data significantly improves the performance of state-of-the-art CNN for SFSU on Foggy Driving; 2) semi-supervised learning further improves performance; and 3) image dehazing marginally advances SFSU with the learning strategy. The datasets, models, and code are publicly available. Keywords: Foggy scene understanding, Semantic segmentation, Object detection, Depth denoising and completion, Dehazing, Transfer learning. The paper introduces an automatic and scalable pipeline to impose high-quality synthetic fog on real clear-weather images, two new datasets (one synthetic and one real) for training and evaluation of models used in SFSU, a new semi-supervised learning approach for SFSU, and a detailed study of the benefit of image dehazing for SFSU and human perception of foggy scenes. The paper is organized as follows: Section 2 presents related work. Section 3 is devoted to the fog simulation pipeline, followed by Section 4 that introduces the two foggy datasets. Section 5 describes supervised learning with synthetic foggy data and studies the usefulness of image dehazing for SFSU in this context. Finally, Section 6 extends the learning to a semi-supervised paradigm, where supervision is transferred from clear-weather images to their synthetic foggy counterparts, and Section 7 concludes the paper. The paper is relevant to image defogging, depth denoising and completion, foggy scene understanding, synthetic visual data, and transfer learning. The authors use the standard optical model for daytime fog to overlay existing clear-weather images with synthetic fog in a physically sound way, simulating the underlying mechanism of foggy image formation. They leverage their fog simulation pipeline to create their Foggy Cityscapes dataset, by adding fog to urbanThis paper addresses the problem of semantic foggy scene understanding (SFSU). Due to the difficulty of collecting and annotating foggy images, the authors generate synthetic fog on real images depicting clear-weather outdoor scenes and use these partially synthetic data for SFSU with state-of-the-art convolutional neural networks (CNN). A complete pipeline is developed to add synthetic fog to real, clear-weather images using incomplete depth information. The authors apply their fog synthesis on the Cityscapes dataset and generate Foggy Cityscapes with 20550 images. SFSU is tackled in two ways: 1) with typical supervised learning, and 2) with a novel type of semi-supervised learning, which combines 1) with an unsupervised supervision transfer from clear-weather images to their synthetic foggy counterparts. Additionally, the usefulness of image dehazing for SFSU is studied. For evaluation, the authors present Foggy Driving, a dataset with 101 real-world images depicting foggy driving scenes, which come with ground truth annotations for semantic segmentation and object detection. Extensive experiments show that 1) supervised learning with synthetic data significantly improves the performance of state-of-the-art CNN for SFSU on Foggy Driving; 2) semi-supervised learning further improves performance; and 3) image dehazing marginally advances SFSU with the learning strategy. The datasets, models, and code are publicly available. Keywords: Foggy scene understanding, Semantic segmentation, Object detection, Depth denoising and completion, Dehazing, Transfer learning. The paper introduces an automatic and scalable pipeline to impose high-quality synthetic fog on real clear-weather images, two new datasets (one synthetic and one real) for training and evaluation of models used in SFSU, a new semi-supervised learning approach for SFSU, and a detailed study of the benefit of image dehazing for SFSU and human perception of foggy scenes. The paper is organized as follows: Section 2 presents related work. Section 3 is devoted to the fog simulation pipeline, followed by Section 4 that introduces the two foggy datasets. Section 5 describes supervised learning with synthetic foggy data and studies the usefulness of image dehazing for SFSU in this context. Finally, Section 6 extends the learning to a semi-supervised paradigm, where supervision is transferred from clear-weather images to their synthetic foggy counterparts, and Section 7 concludes the paper. The paper is relevant to image defogging, depth denoising and completion, foggy scene understanding, synthetic visual data, and transfer learning. The authors use the standard optical model for daytime fog to overlay existing clear-weather images with synthetic fog in a physically sound way, simulating the underlying mechanism of foggy image formation. They leverage their fog simulation pipeline to create their Foggy Cityscapes dataset, by adding fog to urban