FPGA accelerated TinyYOLO v2 object detection neural network, capable of detecting 95 object classes. The design obtained the 5th place out of 65 teams, in the FPGA category, in the System Design Contest in Design Automation Conference 2018, San Fransisco (https://dac.com/content/2018-system-design-contest).
The final rankings are published in http://www.cse.cuhk.edu.hk/~byu/2018-DAC-HDC/ranking.html#final
The team list is in http://www.cse.cuhk.edu.hk/~byu/2018-DAC-HDC/teams.html
The design was deployed in the Xilinx PYNQ-Z1 platform (http://www.pynq.io/)
The design is based on the TinyYOLO v2 Object Detection Neural Network (https://pjreddie.com/darknet/yolo/). We used Half-Precision Floating point (16 bit) our design. The implementation was done on Verilog HDL and using the Vivado 2017.2
The block design of our architecture is as follows,
The Vivado block design connecting our IP to the Zynq Processing System is as follows,
Resource Utilization :
Power estimate :
此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。