Multi-Modal Visibility Improvement under Abnormal Weather Conditions using Contextual Conditional GAN
Maria Siddiqua, Naeem Akhter, Javaid Khurshid · 2021
Removal of multiple weather-induced effects in a single image remains an open problem, although several methods addressing single effect removal have been proposed. We present a single method for removing multiple weather-induced effects that are fog, haze, rain streaks, and snowflakes. The proposed method is a unified model, based on context encoder and conditional generative adversarial network, with a single set of parameters. We demonstrate that our model is effective at improving visibility in weather degraded images. Though the training is done on synthetic data the model generalizes on real images during testing.