ohmybuggg / edgedevicedocker

This is a docker and tutorial for edge device.

und# EdgeDeviceDocker This is a docker and tutorial for edge device.

• ubuntu 20.04 PC
  • Docker version 24.0.1
• Nvidia Jetson AGX Xavier: Ubuntu 20.04
• Jetson Nano 2GB
  • Too low to use cpu.
  • Use LXDE-based desktop to get more RAM.
  • It doesn't have enough RAM to build my docker.

Steps

1. run build.sh to build docker image.
2. run $ xhost + in local to allow host connection.
3. run run.sh to create and start container.

Caution

• Make sure $DISPLAY is the same in local and docker. Run $ echo $DISPLAY to check. If not, set docker's variable to right. Run $ export DISPAY={what's in local}.
• The package version list can't be the same in PC and Xavier, I'm still trying to figure out the problem.
• Dockerfile_xavier is also compatible with PC
• Jetson Nano has just one device, maybe it's due to the ubuntu version.
  • https://unix.stackexchange.com/questions/512759/multiple-dev-video-for-one-physical-device

Todo

• Jetson Nano: 192.168.100.2: Ubuntu 18.04.6 LTS
  • Docker version 20.10.7
  • 1.9G MEM will exploded, so as 100% CPU usage.
  • test_cam.py gives 632M MEM, and 40~50% CPU usage.
  • Dockerfile_cuda can't compatible with nano.
• TX2: Ubuntu 16.04.2 LTS
• Build nvidia/cuda:11.4.3-cudnn8-devel-ubuntu20.04 docker with opencv used CUDA
  • Big failed !!!!!!!!!!!!!!!! -> solved. I am idiot.
  • Docker can run, but GPU can't work with opencv. -> solved

Update

• There is a built docker.
  • plugin-motion-detector
    • everything goes well, just run it.
      • Xavier need to use --runtime=nvidia
    • Can not show display.
  • plugin-objectcounter
    • default use waggle, need to change to local device.
    • at least 8GB remain space -> waggle/plugin-base:1.1.1-ml-torch1.9 this image's file structure isn't same as waggle/plugin-base:1.1.1-base, COPY local in it now.(TBD)
    • nano will crash.
    • CUDA different between PC and edge device (11 vs 10.2) -> update docker didn't work

  • sound-event-detection

Note

• $ docker builder prune to delete docker build cache.

Result table

porject	arch.	platform	status	comments
motion detector	amd64	Ubuntu host	ok	No gpu used in code
	arm64	Jetson Nano	ok	(No gpu used in code) 1. use --runtime=nvidia instead of --gpus all 2. can't show display
	arm64	Jetson Xavier	ok	(No gpu used in code) 1. use --runtime=nvidia instead of --gpus all 2. can't show display
	arm64	Jetson TX2	ok	(No gpu used in code) 1. use --runtime=nvidia instead of --gpus all 2. can't show display
object counter	amd64	Ubuntu host	ok	CUDA 11.4, different image between amd64 and arm64
	arm64	Jetson Nano	failed	CUDA version 10.2 failed to run code, use newer image(CUDA 11.4) will cause another error from opencv
	arm64	Jetson Xavier	failed	CUDA version 10.2 failed to run code, use newer image(CUDA 11.4) will cause another error from opencv
	arm64	Jetson TX2	failed	CUDA version 10.2 failed to run code, use newer image(CUDA 11.4) will cause another error from opencv
sound detector	amd64	Ubuntu host	ok	No cuda used
	arm64	Jetson Nano	ok	No cuda used
	arm64	Jetson Xavier	ok	No cuda used
	arm64	Jetson TX2	ok	No cuda used