The following groups of images are

The following groups of images are the comparison of the proposed two model tests on the real datasets, where (a) is the rainy image, (b) is generated via conditional GAN, (c) is generated via attentive GAN. As we can see from example (1), (c) has the higher contrast compares to (b), and shows better de-raining performance as we can see that the rain streak in left leaves is completely removed in (c), while some streak remained in (b), however, the reflection is also removed, or water droplet, the reason behind it might be that attentive GAN confuses the reflection on the leaves with the raindrop. The according attentive maps show that the attentive network manages to generally move the raindrop focus area from the plant leaves to the real streak.