SegFast-V2: Semantic image segmentation with less parameters in deep learning for autonomous driving
- PDF / 4,860,659 Bytes
- 10 Pages / 595.276 x 790.866 pts Page_size
- 0 Downloads / 253 Views
ORIGINAL ARTICLE
SegFast‑V2: Semantic image segmentation with less parameters in deep learning for autonomous driving Swarnendu Ghosh1 · Anisha Pal2 · Shourya Jaiswal2 · K. C. Santosh3 · Nibaran Das1 · Mita Nasipuri1 Received: 8 February 2019 / Accepted: 21 August 2019 © Springer-Verlag GmbH Germany, part of Springer Nature 2019
Abstract Semantic image segmentation can be used in various driving applications, such as automatic braking, road sign alerts, park assists, and pedestrian warnings. More often, AI applications, such as autonomous modules are available in expensive vehicles. It would be appreciated if such facilities can be made available in the lower end of the price spectrum. Existing methodologies, come with a costly overhead with large number of parameters and need of costly hardware. Within this scope, the key contribution of this work is to promote the possibility of compact semantic image segmentation so that it can be extended to deploy AI based solutions to less expensive vehicles. While developing cheap and fast models one must also not compromise the factor of reliability and robustness. The proposed work is primarily based on our previous model named “SegFast”, and is aimed to perform thorough analysis across a multitude of datasets. Beside “spark” modules and depth-wise separable transposed convolutions, kernel factorization is implemented to further reduce the number of parameters. The effect of MobileNet as an encoder to our model has also been analyzed. The proposed method shows a promising decrease in the number of parameters and significant gain in terms of runtime even on a single CPU environment. Despite all those speedups, the proposed approach performs at a similar level to many popular but heavier networks, such as SegNet, UNet, PSPNet, and FCN. Keywords Compressed encoder–decoder model · Semantic image segmentation · Deep learning
1 Introduction
* K. C. Santosh [email protected] Swarnendu Ghosh [email protected] Anisha Pal [email protected] Shourya Jaiswal [email protected] Nibaran Das [email protected] Mita Nasipuri [email protected] 1
Jadavpur University, Kolkata, India
2
Manipal Institute of Technology, Manipal, India
3
University of South Dakota, Vermillion, USA
Over the last few decades, Artificial Intelligence (AI) has made tremendous progress in various industries. One of the major fields that have received a boost from deep learning is autonomous driving. Many top tier automobile industries have been deploying vehicles with varieties of AI modules, starting from Advanced Driver-Assistance System, such as park assist, anomaly detection and emergency braking, to fully autonomous systems, such as driver-less transportation and auto parking. While the expensive brands and models are aiming for a fully autonomous model, other brands are struggling for the cheaper AI assistive modules/technologies. These technologies must be performed in a restricted hardware configuration without a significa
Data Loading...
