A Deep Motion Deblurring Network based on Per-Pixel Adaptive Kernels with Residual Down-Up and Up-Down Modules, A source code of the 3rd winner of NTIRE 2019 Video Deblurring Challenge
is the local-cov operation position-specific? Assuming that the result's shape is (4,256,256,25,3),the kernel_2d is (4,256,256,25,1),the three channels corresponding to each pixel share kernel_2d's 5×5 parameters?