Auto-tuning for YoloV3

Questions
#1

Hello everyone,

I am running yolov3 using the example tutorial: “Compile YOLO-V2 and YOLO-V3 in DarkNet Models” using RPC and running on pynq-z1. It is taking ~10 seconds to classify an image. I read that you can increase performance by auto-tuning, however, the examples don’t show how to do this with Yolov3.

Could someone provide an example of how to auto-tune a Yolov3 network? I know Yolov3 is supposed to be fast, and I would like to increase my current performance.

Thanks

#2

Take a look at the tune_nnvm tutorial scripts. You should be able to drop in the YOLOv3 graph in place of the other networks in get_network.