How to pass argument to docker image having python script

Question

I am trying to make a classification docker container, but here i am putting a short code. The issue is that docker is not able to take the image folder as input. Here is Dockerfile

FROM pytorch/pytorch:1.6.0-cuda10.1-cudnn7-runtime
RUN apt-get update && apt-get install -y --no-install-recommends \
    # we have found python3.7 in base docker
    python3-pip \
    python3-setuptools \
    build-essential \
    && \
    apt-get clean && \
    python -m pip install --upgrade pip

WORKDIR /workspace
COPY inference.py /workspace
ENTRYPOINT  ["python", "inference.py"]

Here is inference.py file

from glob import glob
import torch
import os

print(os.getcwd())
def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("-i", '--input_folder',  required=True)    
    args = parser.parse_args()
    
    #load data
    images = sorted(glob("{}/*.nii.gz".format(args.input_folder)))
    data_dicts = [
                {"image": image_name}
                for image_name in images
            ]
    print('number of images',len(data_dicts))
    print(torch.cuda.get_device_name(0))
if __name__ == "__main__":
    main()

Using build command i build docker image docker build -t test . This is how i am running docker run --gpus all test -i=images

The folder images is in current directoy. What i am expecting to print the number of images. When i run the image using docker run --gpus all test -i images it print following

/workspace
number of images 0
NVIDIA GeForce GTX 1060

Answer 1

You need to give the container access to your images folder by mapping it into the container filesystem as a volume. You do that using the -v option on the docker run command.

If your images are in the images folder on your computer and the container expects them in the /workspace/images folder, you'd map it like this

docker run --gpus all -v $(pwd)/images:/workspace/images test -i images