상세 보기
초록
Deep-learning technology is widely used in various computer vision tasks. In particular, in a field where high performance is required such as autonomous driving, real-time inference performance of using a highly integrated accelerator is an important. However, frameworks currently provide only vendor-dependent environments, e.g., CUDA, which is problematic in terms of versatility and cost. This paper proposes a general-purpose framework for accelerating object detection performance on real-time video of neural networks based on Open-CL, especially using multi GPUs. Using this approach, the inference speed is faster up to 20 times. ? 2021 IEEE.
키워드
Acceleration; Deep-learning; Multi-GPU; Open-CL; Parallelism; Deep learning; Graphics processing unit; Object recognition; Program processors; Autonomous driving; Detection performance; General purpose framework; Learning technology; Multi-gpu; Real time videos; Real-time inference; Video object detections; Object detection
- 제목
- Open-CL based Multi GPU Acceleration for Video Object Detection
- 저자
- Lee, Young-Woon; Heo, Young-Jin; Cho, Chang-Sik; Kim, Byung-Gyu
- 발행일
- 2021-01
- 유형
- Proceedings Paper
- 저널명
- 2021 IEEE International Conference on Consumer Electronics (ICCE)
- 권
- 2021-January