2020/02/12 by Tomás Völker, Völker, Tomás, Guillaume Boisson +3
Computer Science · Environmental Science · #Advanced Vision and Imaging #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Image Enhancement Techniques #Remote Sensing and LiDAR Applications
paper · pdf · doi:10.48550/arxiv.2002.05028
openalex publication_date 2020/02/12 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
In this paper we address the problem of view synthesis from large baseline\nlight fields, by turning a sparse set of input views into a Multi-plane Image\n(MPI). Because available datasets are scarce, we propose a lightweight network\nthat does not require extensive training. Unlike latest approaches, our model\ndoes not learn to estimate RGB layers but only encodes the scene geometry\nwithin MPI alpha layers, which comes down to a segmentation task. A Learned\nGradient Descent (LGD) framework is used to cascade the same convolutional\nnetwork in a recurrent fashion in order to refine the volumetric representation\nobtained. Thanks to its low number of parameters, our model trains successfully\non a small light field video dataset and provides visually appealing results.\nIt also exhibits convenient generalization properties regarding both the number\nof input views, the number of depth planes in the MPI, and the number of\nrefinement iterations.\n