Implementation of the Pico-CNN Deep Learning Inference Framework in OpenCL for Embedded Devices
Assigned to S. Brandt.
Bachelor’s Thesis / Master’s Thesis / Student Research Project
Abstract
Inference of deep learning algorithms on embedded/edge devices is a very active area of research in academia and industry. However, popular deep learning frameworks are not suited for inference on embedded/edge devices. In order to meet this demand the Chair for Embedded Systems developed the open source deep learning inference framework Pico-CNN [1] (https://github.com/ekut-es/pico-cnn) which is completely written in C, optimized for embedded/edge devices, and is not dependent on third party libraries.
In this student research project an OpenCL [2] variant of Pico-CNN should be implemented and optimized to utilized embedded CPUs and GPUs.
Requirements
- C/C++
- Deep Learning
- Linux (optional)
References
[1] K. Lübeck and O. Bringmann, “A Heterogeneous and Reconfigurable Embedded Architecture for Energy-Efficient Execution of Convolutional Neural Networks,” in Architecture of Computing Systems – ARCS 2019, pp. 267–280 (Copenhagen, Denmark).
[2] Wikipedia, OpenCL: https://en.wikipedia.org/wiki/OpenCL