3 ms·
I've been playing around with the edge TPU on a coral.ai board. It's very impressive - I'm able to do real time object detection at tens of frames per second vs
by throwaway09223 5y ago
I've been playing around with the edge TPU on a coral.ai board. It's very impressive - I'm able to do real time object detection at tens of frames per second vs one frame every 15 or so seconds on my CPU.
- lostmsu 5y agoBTW, this sounded surprising to me, so I went to check it out. According to YOLOv4 GitHub, i7-7700 can only do 4 FPS of detection. For comparison, tuned version on Jetson Xavier (NVidia dev board) does 40FPS. https://github.com/AlexeyAB/darknet https://github.com/AlexeyAB/darknet https://github.com/indra4837/yolov4_trt_ros https://github.com/indra4837/yolov4_trt_ros
- throwaway09223 5y agoYes, that sounds about right. I was using YOLOv3 on a i7-1065G7 (laptop) which probably explains the 4 FPS vs 0.25 FPS difference on CPU performance. I put the project down for a while because I got busy. I really should revive it.