TY - GEN
T1 - Synthetic-to-Real Domain Adaptation for Lane Detection
AU - Garnett, Noa
AU - Uziel, Roy
AU - Efrat, Netalee
AU - Levi, Dan
N1 - Publisher Copyright:
© 2021, Springer Nature Switzerland AG.
PY - 2021/1/1
Y1 - 2021/1/1
N2 - Accurate lane detection, a crucial enabler for autonomous driving, currently relies on obtaining a large and diverse labeled training dataset. In this work, we explore learning from abundant, randomly generated synthetic data, together with unlabeled or partially labeled target domain data, instead. Randomly generated synthetic data has the advantage of controlled variability in the lane geometry and lighting, but it is limited in terms of photo-realism. This poses the challenge of adapting models learned on the unrealistic synthetic domain to real images. To this end we develop a novel autoencoder-based approach that uses synthetic labels unaligned with particular images for adapting to target domain data. In addition, we explore existing domain adaptation approaches, such as image translation and self-supervision, and adjust them to the lane detection task. We test all approaches in the unsupervised domain adaptation setting in which no target domain labels are available and in the semi-supervised setting in which a small portion of the target images are labeled. In extensive experiments using three different datasets, we demonstrate the possibility to save costly target domain labeling efforts. For example, using our proposed autoencoder approach on the llamas and tuSimple lane datasets, we can almost recover the fully supervised accuracy with only 10% of the labeled data. In addition, our autoencoder approach outperforms all other methods in the semi-supervised domain adaptation scenario.
AB - Accurate lane detection, a crucial enabler for autonomous driving, currently relies on obtaining a large and diverse labeled training dataset. In this work, we explore learning from abundant, randomly generated synthetic data, together with unlabeled or partially labeled target domain data, instead. Randomly generated synthetic data has the advantage of controlled variability in the lane geometry and lighting, but it is limited in terms of photo-realism. This poses the challenge of adapting models learned on the unrealistic synthetic domain to real images. To this end we develop a novel autoencoder-based approach that uses synthetic labels unaligned with particular images for adapting to target domain data. In addition, we explore existing domain adaptation approaches, such as image translation and self-supervision, and adjust them to the lane detection task. We test all approaches in the unsupervised domain adaptation setting in which no target domain labels are available and in the semi-supervised setting in which a small portion of the target images are labeled. In extensive experiments using three different datasets, we demonstrate the possibility to save costly target domain labeling efforts. For example, using our proposed autoencoder approach on the llamas and tuSimple lane datasets, we can almost recover the fully supervised accuracy with only 10% of the labeled data. In addition, our autoencoder approach outperforms all other methods in the semi-supervised domain adaptation scenario.
UR - http://www.scopus.com/inward/record.url?scp=85103247042&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-69544-6_4
DO - 10.1007/978-3-030-69544-6_4
M3 - Conference contribution
AN - SCOPUS:85103247042
SN - 9783030695439
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 52
EP - 67
BT - Computer Vision – ACCV 2020 - 15th Asian Conference on Computer Vision, 2020, Revised Selected Papers
A2 - Ishikawa, Hiroshi
A2 - Liu, Cheng-Lin
A2 - Pajdla, Tomas
A2 - Shi, Jianbo
PB - Springer Science and Business Media Deutschland GmbH
T2 - 15th Asian Conference on Computer Vision, ACCV 2020
Y2 - 30 November 2020 through 4 December 2020
ER -