Deep Generative Model for Single Image De-Hazing on Embedded Platform

Haze can severely undermine the visibility and can often deteriorate the performance of computer vision applications like object detection, object classification etc. Hence, image de-hazing is a very desirable task. The existing state of the art methods perform well on qualitative measures but often overlook memory efficiency, which is especially crucial when working with embedded platforms, especially for realistic applications. The proposed method in this paper utilises a deep generative network that incorporates Wasserstein loss which significantly improves the quality of images. Moreover; the proposed method is parameter efficient and utilises nearly 50 times lesser parameters than the existing state of the art deep generative models which have been used for de-hazing.