Deep learning has revolutionized vision sensing applications in terms of accuracy comparing to other techniques. Its breakthrough comes from the ability to extract complex high level features directly from sensor data. However, deep learning models are still yet to be natively supported on mobile devices due to high computational requirements. In this paper, we present DeepMon, a next generation of DeepSense  framework, to enable deep learning models on conventional mobile devices (e.g. Samsung Galaxy S7) for continuous vision sensing applications. Firstly, Deep-Mon exploits similarity between consecutive video frames for intermediate data caching within models to enhance inference latency. Secondly, DeepMon leverages approximation technique (e.g. Tucker decomposition) to build up approximated models with negligible impact on accuracy. Thirdly, DeepMon ofloads heavy computation onto integrated mobile GPU to significantly reduce execution time of the model.
Continuous vision, Deep learning, Mobile GPU, Mobile sensing
Hardware Systems | Software Engineering
Software and Cyber-Physical Systems
MobiSys 2017: Proceedings of the 15th Annual International Conference on Mobile Systems, Applications, and Services, Niagara Falls; United States, 2017 June 19-23
Taylor & Francis: STM, Behavioural Science and Public Health Titles
City or Country
Niagara Falls; United States
HUYNH, Nguyen Loc; BALAN, Rajesh Krishna; and LEE, Youngki.
DEMO: DeepMon - Building mobile GPU Deep learning models for continuous vision applications. (2017). MobiSys 2017: Proceedings of the 15th Annual International Conference on Mobile Systems, Applications, and Services, Niagara Falls; United States, 2017 June 19-23. Research Collection School Of Information Systems.
Available at: http://ink.library.smu.edu.sg/sis_research/3672
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.