Depth Estimation and Image Restoration by Deep Learning from Defocused Images

Monocular depth estimation and image deblurring are two fundamental tasks in computer vision, given their crucial role understanding 3D scenes. Performing any of them by relying on a single is an ill-posed problem. The recent advances the field Deep Convolutional Neural Networks (DNNs) have revolutionized many including deblurring. When it comes to using defocused images, recovery All-in-Focus (Aif) become related problems due defocus physics. Despite this, most existing models treat separately. There are, however, that solve these simultaneously concatenating networks sequence first estimate or map then reconstruct focused based it. We propose DNN solves parallel. Our Two-headed Depth Estimation Deblurring Network (2HDED:NET) extends conventional from Defocus (DFD) with branch shares same encoder as branch. proposed method has been successfully tested benchmarks, one for indoor other outdoor scenes: NYU-v2 Make3D. Extensive experiments 2HDED:NET benchmarks demonstrated superior close performances those state-of-the-art