A new approach for training a physics-based dehazing network using synthetic images

•We present a novel method of using synthetic images, captured from a 3D urban virtual environment, for a physics-based dehazing task.•A style transfer strategy, followed by unlit image prior extraction from the synthetic images, allows the dehazing network to perform effectively on real-world hazy images.•A three-stage dehazing network is presented, which is easier to train as compared to unsupervised approaches.•Our method achieves competitive performance with other state-of-the-art methods, without the network seeing any real hazy images during training.•We provide DLSU-SYNSIDE (SYNthetic Single Image Dehazing) dataset, which contains 100K synthetic images.