How to Run Tensorflow Object Detection with Docker on Jetson Modules?

How to Run Tensorflow Object Detection with Docker on Jetson Modules?

Jetson AGX Xavier | Jetson Nano | Jetson TX2 NX | Jetson Xavier NX

06 April 2021

1- How to pull Docker Image from Nvidia NGC

2- How to configure Docker Container with using graphic interface

3- How to test TensorFlow environment with custom masked face detection model


Hardware: DSBOX-NX2

OS: JetPack 4.5 (L4T-32.5)

In this blog post, we are going to explain how to run TensorFlow Object Detection with Docker on Jetson Modules. The process will be the same for all Jetson Modules. This GitHub repo has been taken as a reference for the whole process.

Firstly, we must pull our docker image from this link. We will be using "NVIDIA L4T ML" image. You can just pull the image by typing the following command!

docker pull

For creating a Docker container, you should be typing the following command. We have passed display parameters because of getting image after object detection and we have passed "-rm" parameter for removing the container after exit the container.

sudo docker run -it --rm --gpus all -e DISPLAY=:0 -v /tmp/.X11-unix:/tmp/.X11-unix –runtime nvidia --network host

Now, we are inside of the container. We should clone the TensorFlow model repository like below.

apt update
apt-get install git
git clone

We should change the working directory.

cd models/research/object_detection/

Then, we need to pull the TensorFlow object detection API repository like below.

git clone

We need to copy the file of TensorFlow object detection API repository inside the "/models/research/object_detection" folder like below.

cp -r TensorFlow-Object-Detection-API-Tutorial-Train-Multiple-Objects-Windows-10/* .
rm -rf TensorFlow-Object-Detection-API-Tutorial-Train-Multiple-Objects-Windows-10/

We need to change the python version that we will be using for object detection.

alias python=python3

We have used the mask model which is trained before. If you would like to train your custom model, you can check this blog for training custom object detection model with TensorFlow.

You can find the trained model labelmap at the top of the post. You must get these models and label map inside the docker container.


We must install unzip and use command on terminal to extract the zip file.  

apt install unzip

We change the name of the files in the folder extracted.

mv frozen_inference_graph.pb ./inference_graph/
mv labelmap.pbtxt ./training/

Download the video that you can use with ‘wget’ command like below. (The Video (video.mp4) link is available at the top of this post.)


We have also changed the using video name on "" file.

sed -i ‘s/’

Please pay attention to model, labelmap and video directories for avoiding possible problems.

On our NVIDIA machine (outside docker container), we need to give the access permission to everyone for avoiding display problems.

Open a new terminal and use ‘xhost +’ command.

xhost +

Then we turn back to the first terminal that we use.

After modifying the model files, we are ready to run to detection on image.

cd ..
protoc object_detection/protos/*.proto --python_out=.
cd object_detection/

You can see the screenshots below.

You can customize the detection image by changing the image name and path in "" file. You can also run to detect model on images and webcam display. 

Thank you for reading our blog post. 

If you want to get updated on new blog posts, product launches and discounts, you can fill out the form and sign up for our newsletter. By signing up, you can reach various blog posts about AI, deep learning, machine vision, high-speed cameras, and industrial interfaces. 

You will receive a "free shipping" code for next purchase immediately.