grannycola / support_detection

Important

1. This project uses git lfs. You will need to:

• clone this repository
• change dir to repo dir
• git lfs pull

2. Install mmdetection and mmcv

How to start train/val process:

1. Specify my_config.py for your device.

2. Run:

python /home/$USERNAME/mmdetection/tools/train.py my_config.py

or specify your mmdetecion path.

Dockerfile

Required:

• Nvidia GPU
• Nvidia Driver.
• Cuda 12.1.0
• nvidia-container-runtime

Check file content /etc/docker/daemon.json It should look like:

{
  "runtimes": {
    "nvidia": {
      "path": "/usr/bin/nvidia-container-runtime",
      "runtimeArgs": []
    }
  },
  "default-runtime": "nvidia"
}

Build image:

sudo docker build -t support_detection .

Run container:

sudo docker run --memory=16g --gpus all -it support_detection

Run training in docker bash:

python /mmdetection/tools/train.py my_config_docker.py

Tested on Nvidia RTX 3090Ti with Cuda driver 12.1.0