Flexible platform for embedded computer vision application using deep learning and GPUs | |
| Author | Wickramatilake, Tharaka Theekshana |
| Call Number | AIT Thesis no.CS-18-02 |
| Subject(s) | Graphics processing unit Engineering graphics |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Computer Science, School of Engineering and Technology |
| Publisher | Asian Institute of Technology |
| Series Statement | Thesis ; no. CS-18-02 |
| Abstract | This thesis will try to address the issue of requiring full size computers for applications whose compute requirements lead to a need for graphics processing unit (GPU) intensive processing needs. GPUs are powerful devices that enable users to execute parallel workloads. Currently, libraries such as CUDA and OpenCL are becoming useful for general purpose computing. Workloads based on deep neural network frameworks or computer vision algorithms such as Farneback Optical Flow even on a modern i7 desktop based processor show insufficient performance for real time processing. Running such workloads on embedded systems with ARM processors has not been possible up to now due to their low performance. Currently, production systems using GPUs use desktop or server class computers, which involve their own set of deployment challenges, such as the size of the system and high power draw. Recently, many companies have developed embedded systems with GPUs that support OpenCL for parallel computing. But these systems lack CUDA support, and almost every popular neural network framework uses CUDA for computing. In order to close the gap between high performance GPU computing with CUDA and current lineups of embedded systems, Nvidia has recently introduced an embedded system module consisting of an Nvidia GPU and a ARM 64bit processor. This research will focus on using this device to provide solutions to myriad applications requiring the above mentioned resources, especially in the area of computer vision |
| Year | 2018 |
| Corresponding Series Added Entry | Asian Institute of Technology. Thesis : no. CS-18-02 |
| Type | Thesis |
| School | School of Engineering and Technology (SET) |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Computer Science (CS) |
| Chairperson(s) | Dailey, Matthew N.; |
| Examination Committee(s) | Mongkol Ekpanyapong;Abeykoon, A.M. Harsha S.; |
| Scholarship Donor(s) | Asian Institute of Technology Fellowship; |
| Degree | Thesis (M. Eng.) - Asian Institute of Technology, 2018 |