Please use this identifier to cite or link to this item:
Title: T2Net : synthetic-to-realistic translation for solving single-image depth estimation tasks
Authors: Zheng, Chuanxia
Cham, Tat-Jen
Cai, Jianfei
Keywords: Engineering::Computer science and engineering
Issue Date: 2018
Source: Zheng, C., Cham, T.-J., & Cai, J. (2018). T2Net : synthetic-to-realistic translation for solving single-image depth estimation tasks. Computer Vision – ECCV 2018, 798-814. doi:10.1007/978-3-030-01234-2_47
Abstract: Current methods for single-image depth estimation use training datasets with real image-depth pairs or stereo pairs, which are not easy to acquire. We propose a framework, trained on synthetic image-depth pairs and unpaired real images, that comprises an image translation network for enhancing realism of input images, followed by a depth prediction network. A key idea is having the first network act as a wide-spectrum input translator, taking in either synthetic or real images, and ideally producing minimally modified realistic images. This is done via a reconstruction loss when the training input is real, and GAN loss when synthetic, removing the need for heuristic self-regularization. The second network is trained on a task loss for synthetic image-depth pairs, with extra GAN loss to unify real and synthetic feature distributions. Importantly, the framework can be trained end-to-end, leading to good results, even surpassing early deep-learning methods that use real paired data.
ISBN: 9783030012335
DOI: 10.1007/978-3-030-01234-2_47
Rights: © 2018 Springer Nature Switzerland AG. All rights reserved. This paper was published in Computer Vision – ECCV 2018 and is made available with permission of Springer Nature Switzerland AG.
Fulltext Permission: open
Fulltext Availability: With Fulltext
Appears in Collections:IMI Conference Papers

Citations 50

Updated on Mar 2, 2021

Citations 50

Updated on Mar 2, 2021

Page view(s)

Updated on Mar 4, 2021

Download(s) 50

Updated on Mar 4, 2021

Google ScholarTM




Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.