Fast Spatio-Temporal Multi-Branch Fusion Network for Video Deraining

High-quality videos are essential for outdoor vision systems. Rain is one of the most familiar factors for degrading video quality. Existing works tend to design the complex network architectures with lots of parameters. Unfortunately, they always generate the nonideal results with residual rain streaks and insufficient details, although the network parameters are ample. To handle these issues, we develop a fast Spatio-Temporal Multi-Branch Fusion Network (STMBFN) for effectively handling video deraining. To be concrete, we design a temporal information generation module to align different frames. Different from existing works to directly concatenate these aligned frames, we divide these frames to ensure the closest relationship between the current and other frames. In this way, we can weaken the burden of the network. As a result, we design a multi-branch fusion network based on three three-layers basic networks. In the experimental part, we present sufficient analyses in terms of our STMBFN to illustrate the effectiveness. Further, plenty of experiments are conducted to fully verify our superiority against existing state-of-the-art approaches.