Installing detectron2 in 9 easy steps

Image generated using Bing

When installing detectron2, we often face an error saying that there is a mismatch between the CUDA version with which PyTorch (say, 11.7) was compiled and the CUDA version available in the system (say, 12.1).

A possible solution is to use a Docker container

In this article, we will go through the steps to set up a Docker container inside the Windows Subsystem for Linux (WSL2) on Windows 11.

#Step-by-Step Installation

#Step 1: Install WSL2 on Windows

Install WSL2 on WindowsFollow the steps givenhere.

#Step 2: Install Docker Desktop

Install Docker Desktop on WindowsFollow the steps givenhere.

#Step 3: Pull CUDA Docker Image

docker pull nvcr.io/nvidia/cuda:11.7.0-base-ubuntu22.04

#Step 4: Run the Docker Container

docker run -it --gpus all nvcr.io/nvidia/cuda:11.7.0-base-ubuntu22.04

#Step 5: Update and Upgrade Packages

apt-get update && apt-get upgrade

#Step 6: Install Python 3.x

apt install python3

Seventh: Install pip because you will need it

apt install python3-pip

Eight: Install the required libraries

For opencv dependencies, run the following command first:

apt-get install ffmpeg libsm6 libxext6

otherwise, when importing detectron2, an ImportError like

ImportError: libGL.so.1: cannot open shared object file: No such file or directory

will occur. If you install libgl1 after seeing this error, you will get

ImportError: libgthread-2.0.so.0: cannot open shared object file: No such file or directory

when you try to import detectron2 again. I did not face any issues when installing opencv, but when importing detectron2 later, the above errors occurred.

To install the required libraries, run

python3 -m pip install numpy matplotlib scikit-learn scikit-image opencv-python opencv-contrib-python …

To install PyTorch, run

python3 -m pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu117

Ninth: Install detectron2

First, install git

apt install git

Then install detectron2

python3 -m pip install 'git+https://github.com/facebookresearch/detectron2.git'

Save this Container as an Image

Run

docker ps

to check the container ID. To commit the changes to the container, run

docker commit [CONTAINED ID] [REPOSITORY[:TAG]]

In my case, the command was

docker commit 870b62170e7e detectron2_cuda11.7.0-ubuntu22.04

Check the newly created Image by running

docker images

Stop the running container and start the new image

docker run -it --gpus all detectron2_cuda11.7.0-ubuntu22.04

There you go! Now the installed libraries won’t vanish every time you remove the container and can be used on other systems as well!

Additional Features:

NVIDIA Driver

Using

--gpus=all

is essential; otherwise, the following error will occur:

AssertionError: Found no NVIDIA driver on your system. 
Please check that you have an NVIDIA GPU and 
installed a driver from http://www.nvidia.com/Download/index.aspx

To access a directory on the host system from within the docker container use bind mounts

docker run -it --gpus all --mount type=bind,src=/path/to/directory,target=/path detectron2_cuda11.7.0-ubuntu22.04

Now, you can access all the files in the source directory on the host system.

To enable the nvidia-smi tool in the container, enable the utility driver capability

docker run -it --gpus 'all,capabilities=compute,utility' --mount type=bind,src=/path/to/directory,target=/path detectron2_cuda11.7.0-ubuntu22.04

Clap and share if you like this article. Subscribe for more articles like this.

Originally published on Medium.