Alona Baruhov


Viterbi Faculty of Electrical Engineering, Technion

Cyclic Generative Models for Depth Image Enhancement

Deep generative models have demonstrated great success in recent years in producing detailed and visually appealing images, achieving breakthrough results in a wide variety of computer vision applications. One of the most influential contributions to such generative models was the Cycle-GAN. Our work focuses on the application of the Cycle-GAN framework to depth image generation and enhancement. Specifically, we aim to generate and enhance depth images provided by a real-world depth sensor, for which an analytic noise model is not available.  We focus on the unsupervised problem, where no ground truth image pairs are at hand. Instead, we formulate it as a domain-transfer problem between low-quality and high-quality sensor domains. Using a dataset of unpaired depth images captured in a free manner by two sensors, based on different technologies, we design and train a translation network which is able to bring the low-quality depth images near the high-quality images. *This seminar is part of a M.Sc. final paper supervised by Prof. Guy Gilboa.

Date: Wed 20 Feb 2019

Start Time: 11:30

End Time: 12:30

1061 | Electrical Eng. Building