Blur-Invariant Deep Learning for Blind-Deblurring ( Supplementary Material )

We provide additional quantitative and qualitative comparisons on two more publicly available datasets, Köhler et al. [2] and Lai et al. [3] respectively, in this supplementary material to demonstrate the efficacy of our proposed network. We also provide an analysis on the deblurring efficiency and generalizability of our network when compared to a network learned for a single blur kernel. Along with this, we also perform qualitative and quantitative study on the blur kernels. We use least squares to estimate the blur kernel from blur/deblurred pair generated from our network and use a kernel similarity metric to quantify the performance.