Robust Attention Deraining Network for Synchronous Rain Streaks and
Raindrops Removal
Abstract
Synchronous Rain streaks and Raindrops Removal (SR3) is a very hard and
challenging task, since rain streaks and raindrops are two wildly
divergent real-scenario phenomena with different optical properties and
mathematical distributions. As such, most of existing deep
learning-based Singe Image Deraining (SID) methods only focus on one of
them or the other. To solve this issue, we propose a new, robust and
hybrid SID model, termed Robust Attention Deraining Network (RadNet)
with strong robustenss and generalztion ability. The robustness of
RadNet has two implications :(1) it can restore different degenerations,
including raindrops, rain streaks, or both; (2) it can adapt to
different data strategies, including single-type, superimposed-type and
blended-type. Specifically, we first design a lightweight robust
attention module (RAM) with a universal attention mechanism for coarse
rain removal, and then present a new deep refining module (DRM) with
multi-scales blocks for precise rain removal. The whole process is
unified in a network to ensure sufficient robustness and strong
generalization ability. We measure the performance of several SID
methods on the SR3 task under a variety of data strategies, and
extensive experiments demonstrate that our RadNet can outperform other
state-of-the-art SID methods.