YoloV4- Install and run Yolo on NVIDIA Jetson Xavier NX with use of GPU

In this tutorial I explain step by step how to install and run YoloV4 on NVIDIA Jetson Xavier NX platform by using its 384 NVIDIA CUDA cores.

For more information:

[arslanMCL]

I tried your examples and it works but getting frame rate of 3 FPS for Tiny and 1.4 for Main example. How I can speed up the results.

[ShyamD]

i have a doubt, as Nvidia has already modified tensorflow as tensor RT, can we make use of it in backend of YOLO ?

[thefirstoct]

Can a new network be trained on Jetson Xavier NX or AGX xavier?

[jeffhan]

Great video! In the yolo-tiny section, why the GPU is not using 99% power while the frame rate is only 10 ? (I thought the algorithm would try to reach 30 fps at any cost)

[haniffjunos]

Hello thank you for the good tutorial...i managed to follow the tutorial up until running the darknet detector demo command.. i go this error: "videoio error: v4l: can't open camera by index 0". this has caused the video stream to stop.
DO you know how to solve this?

[vaibhavp.]

Hello Sir, I want to download a specific version of the OpenCV on my Jetson TX2 and also want to downgrade the CUDA version to run a GitHub repository. Can you tell me how can I do that on my board without re-flashing the Jetson Tx2 board?. Thanks a lot.

[yashsonavane]

Will you help me for same thing for jetson nano.... Cause I am not able to activate cuda on jetson nano

[dhanyagopinath]

your video is really inspiring. Can we do the same on Jetson Nano.

[mokhtarhafez]

Are you using an M.2 or running directly from an SD card ?, because it seems that I am getting stuck ok 3.7 FPS regardless of the yolo version

[mokhtarhafez]

I have ran everything, but I seem to be stuck on an average of 3.7fps on yolov4-tiny,

I am using jetpack4.6 with cuda 10.2 and I have recompiled open cv with cuda, but it seems that the fps is capped somehow, what could be a possible solution

[yashsonavane]

I am using Yolov4-deepsort code it's not at all utilizing GPU...
Is there any solution for that

[jinhwalee]

Great Video :) If you can check the reply, can i ask to something? How did you do recording your screen? remote program..or other tools? plz!

[이동규-qp]

Hi, I have a question. I've been all set until compiling with the 'make',
but when I tried the testing with 'darknet detector demo cfg/coco.data cfg/yolov4.cfg yolov4.weights c 0', OR
'./darknet detector demo cfg/coco.data cfg/yolov4 tiny.cfg yolov4 tiny.weights -c 0', with the v4 and v4tiny I get 3 errors of

1. [WARN:0] global ~~~~(1757) / Gstreamer warning: Embedded video playback halted; module v4l2src0 reported: Internal data stream error.
2. [WARN:0] global ~~~~(886) / GStreamer warning: unable to start pipeline
3. [WARN:0] global ~~~~(480) / Gstreamer warning: GStreamer: pipeline has not been created
With the v4, I get continuous lines of VIDIOC_QBUF: Invalid argument coming up
With the v4tiny, I have FPS of 4.0, but with Object:___ (empty). A demo screen shows up but filled with a green screen with no change.

Is there something I have missed? What I haven't done is the github AlexeyAB/darknet part, because I did not quite understand what should be done.
Another anticipation is about the size of the Video Stream. I found out that I was getting different video stream sizes. I am trying to find out how to change it, but I'm stuck :(

Thank you for great video!