SAAFNet: Self-Adaptive Attention Fusion Network for Single Image Deraining

Adverse weather conditions, particularly raindrops on camera lenses, significantly degrade the accuracy of AI models for outdoor cameras. To address this issue, we propose SAAFNet, a novel U-Net-based architecture designed for single image deraining. SAAFNet employs a novel Self-Adaptive Attention Fusion (SAAF) block, Depthwise Separable Convolutions, and an attention mechanism to achieve high-fidelity image restoration while maintaining a lightweight structure suitable for real-time applications. Experimental results demonstrate SAAFNet’s superior deraining performance (PSNR: 26.82 dB, SSIM: 0.89) and significantly reduced computational footprint (FLOPs: 21.83G, Params: 4.13M) compared to state-of-the-art methods, enhancing downstream task accuracy.