Name: Csaba Benedek
Affiliation: HUN-REN Institute for Computer Science and Control (HUN-REN SZTAKI)
Primary research interest: Analysis of geospatial data, Lidar point cloud based 3D object recognition and reconstruction, Bayesian classification and object modelling approaches in computer/machine vision applications
Title of the lecture: Generative Methods on 2D Image and 3D Structure Inpainting
Keywords: GAN, image inpainting, shape completion, Lidar
Summary: In this lecture, novel deep learning-based methods are presented for automatic analysis and inpainting of 2D images and 3D structures. The proposed methods are focused on three specifc use cases:
(i) inpainting 2D images of masonry walls in archaeology, and civil engineering applications,
(ii) filling up the missing regions in point clouds of 3D models acquired using Mobile Laser Scanning (MLS) systems. (
iii) dense depth image prediction from sparse meaurements of a cost-efficient Lidar sensor.
The proposed methods are evaluated on large databases containing a variety of synthetic and real-world scenarios, and their advantages over recent state-ofthe-art approaches are demonstrated.