MSSNet: Multi-Scale-Stage Network for Single Image Deblurring

Most traditional single image deblurring methods before deep learning adopt a coarse-to-fine scheme that estimates sharp at coarse scale and progressively refines it finer scales. While this has also been adopted in several learning-based approaches, recently number of single-scale approaches have introduced showing superior performance to previous terms quality computation time. In paper, we revisit the analyze defects approaches. Based on analysis, propose Multi-Scale-Stage Network (MSSNet), novel approach with our remedies defects. MSSNet adopts three remedies: stage configuration reflecting blur scales, an inter-scale information propagation scheme, pixel-shuffle-based multi-scale scheme. Our experiments show can effectively resolve improve performance.