DeRaindrop

Why we included this project

DeRaindrop is the reference implementation of a CVPR 2018 paper on a real imaging problem: raindrops on a window or lens hide the scene behind them, and this code removes them from a single image. The model is an attentive GAN that feeds visual attention into both the generator and the discriminator, so the generator concentrates on the raindrop regions and their surroundings while the discriminator checks local consistency of the restored areas. That makes it directly relevant to outdoor camera work, autonomous driving perception, and surveillance, where weather-degraded frames are a constant nuisance. The code targets Python 2.7 with PyTorch and CUDA, so treat it as a research baseline to study and adapt rather than something you would drop into production. It is also a handy reference if you are exploring attention mechanisms inside GANs for low-level vision.