Large scale outdoor RGB+D dataset and its application to single image depth estimation
*bldg. 2 NO.102/ 16:00~
Dongbo Min received the BS, MS, and PhD degrees from the School of Electrical and Electronic Engineering, Yonsei University, Seoul, South Korea, in 2003, 2005, and 2009, respectively. He was a post-doctoral researcher in the Mitsubishi Electric Research Laboratories, Cambridge, Massachusetts, from 2009 to 2010. From 2010 to 2015, he was with the Advanced Digital Sciences Center, Singapore. From 2015 to 2018, he was an assistant professor in the Department of Computer Science and Engineering, Chungnam National University, Daejeon, South Korea. Since 2018, he has been an assistant professor in the Department of Computer Science and Engineering, Ewha Womans University, Seoul, South Korea. His current research interests include computer vision, 2D/3D video processing, computational photography, augmented reality, and continuous/discrete optimization. He is a senior member of the IEEE.
Obtaining 3D depth of a scene is essential to alleviate a number of challenges in computer vision tasks. A great variety of computational stereo approaches have been proposed to infer a depth map when two (or more) images are given as inputs. Recently, monocular depth estimation that aims to estimate the depth map from a single image has been studied actively thanks to the widespread of deep neural networks. This talk introduces our recent works for the monocular depth estimation, including constructing a large-scale RGB+D dataset and novel convolutional neural networks based algorithms. We first present the massive RGB+D dataset, called DIML/CVL dataset, which provides 1M color images and associated depth maps taken in outdoor environments. This includes several pre-processing algorithms and detailed performance analysis of our dataset. Then, we introduce a novel approach for monocular depth estimation based on a deep variational model.