Attentive Generative Adversarial Network for Raindrop Removal from a Single Image
share
Raindrops adhered to a glass window or camera lens can severely hamper the visibility of a background scene, and degrade an image considerably. In this paper, we address the problem by visually removing raindrops, and thus transforming a raindrop degraded image into a clean image. The problem is intractable, since first which regions are occluded by raindrops are not given. Second, the information about the background scene of the occluded regions for most part is completely lost. To resolve the problem, we apply an attentive generative network using the idea of adversarial training. Our main idea is to inject visual attention into both the generative and discriminative networks. In the training stage, our visual attention is guided by the locations of raindrop regions. Hence, by injecting this, the generative network will pay more attention to the raindrop regions and their surroundings which are the regions we want to modify; and, the discriminative network will be able to assess the local consistency of the restored regions. To our knowledge, this injection of visual attention to both generative and discriminative networks is novel. Our experiments show the effectiveness of our approach, which outperforms the state of the art methods quantitatively and qualitatively.
READ FULL TEXT