FMD-cGAN: Fast Motion Deblurring Using Conditional Generative Adversarial Networks

AbstractIn this paper, we present a Fast Motion Deblurring-Conditional Generative Adversarial Network (FMD-cGAN) that helps in blind motion deblurring of single image. FMD-cGAN delivers impressive structural similarity and visual appearance after an Like other deep neural network architectures, GANs also suffer from large model size (parameters) computations. It is not easy to deploy the on resource constraint devices such as mobile robotics. With help MobileNet [1] based architecture consists depthwise separable convolution, reduce inference time, without losing quality images. More specifically, by 3–60x compare nearest competitor. The resulting compressed Deblurring cGAN faster than its closest competitors even qualitative quantitative results outperform various recently proposed state-of-the-art models. We can use our for real-time image tasks. current experiment standard datasets shows effectiveness method.KeywordsFast deblurringGenerative adversarial networksDepthwise convolutionHinge loss