Real-Time Monocular Depth Estimation using Synthetic Data with Domain Adaptation via Image Style Transfer

Atapour-Abarghouei, A.; Breckon, T.P.

Abstract

Monocular depth estimation using learning-based approaches has become promising in recent years. However, most monocular depth estimators either need to rely on large quantities of ground truth depth data, which is extremely expensive and difficult to obtain, or predict disparity as an intermediary step using a secondary supervisory signal leading to blurring and other artefacts. Training a depth estimation model using pixel-perfect synthetic data can resolve most of these issues but introduces the problem of domain bias. This is the inability to apply a model trained on synthetic data to real-world scenarios. With advances in image style transfer and its connections with domain adaptation (Maximum Mean Discrepancy), we take advantage of style transfer and adversarial training to predict pixel perfect depth from a single real-world color image based on training over a large corpus of synthetic environment data. Experimental results indicate the efficacy of our approach compared to contemporary state-of-the-art techniques.

Citation

Atapour-Abarghouei, A., & Breckon, T. (2018, June). Real-Time Monocular Depth Estimation using Synthetic Data with Domain Adaptation via Image Style Transfer. Presented at 2018 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)., Salt Lake City, Utah, USA

• Presentation Conference Type: Conference Paper (published)
• Conference Name: 2018 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).
• Start Date: Jun 18, 2018
• End Date: Jun 22, 2018
• Acceptance Date: Feb 19, 2018
• Online Publication Date: Dec 17, 2018
• Publication Date: 2018
• Deposit Date: Mar 19, 2018
• Publicly Available Date: Mar 20, 2018
• Pages: 2800-2810
• Series ISSN: 2575-7075
• Book Title: Proc. Computer Vision and Pattern Recognition
• DOI: https://doi.org/10.1109/CVPR.2018.00296
• Keywords: monocular depth, generative adversarial network, GAN, depth map, disparity, depth from single image, style transfer
• Public URL: https://durham-repository.worktribe.com/output/1145708
• Publisher URL: https://breckon.org/toby/publications/papers/abarghouei18monocular.pdf