Inverted Residual Fourier Transformation for Lightweight Single Image Deblurring

Single image deblurring aims to restore a sharp by removing blurred areas in the single image. Such images are not only visually unpleasant, but cause various problems many applications like recognition. In recent years, with development of deep learning, neural networks used deblurring. Especially, encoder-decoder structures widely for and successfully high quality images. However, FLOPs number parameters tend increase high-quality Thus, this paper proposes new lightweight network (IRFTNet) based on UNet, which is basic structures. Our proposed has three features improve performance lightweight. First, backbone called Inverted Residual Fourier Transformation block (IRFTblock) inverted residual introduced decrease computational complexity. Second, module Lower Feature Synthesis (LFS) efficiently transfer encoder information from lower layers upper layers. Finally, multiple outputs structure MIMO-UNet introduced. These improvements resulted 32.98dB PSNR GoPro dataset, despite approximately half DeepRFT. Further ablation studies show effectiveness components our model.