This thesis investigates possible deep learning-based solutions for three challenging computer vision problems. We first tackle depth from focus and devise a network architecture for it. Following, we present a fusion-based CNN architecture to incorporate depth into semantic segmentation. Furthermore, we propose a multimodal CNN architecture that exploits pixelwise semantic labels in addition to color to improve the image restoration tasks. Consequently, we discuss the limitations and provide directions for future work.
«
This thesis investigates possible deep learning-based solutions for three challenging computer vision problems. We first tackle depth from focus and devise a network architecture for it. Following, we present a fusion-based CNN architecture to incorporate depth into semantic segmentation. Furthermore, we propose a multimodal CNN architecture that exploits pixelwise semantic labels in addition to color to improve the image restoration tasks. Consequently, we discuss the limitations and provide di...
»