Introduction

This is a sand box repository to use the MMDetection library which is an open source object detection toolbox based on PyTorch. It is a part of the OpenMMLab project.

Environment Details

• Windows 11 WIP build 22504.1010
• WSL2
  • Kernel 5.10.60.1
  • Ubuntu 20.04.3 LTS (Installed from windows store)
• CUDA
  • Toolkit 11.3
  • NvidiaDriver version 472.34
  • Nvidia CUDA Driver 11.4.150
• Hardware
  • Laptop: Lenovo X1 Xtreme
  • CPU: Intel Core i7-9750H
  • GPU: NVIDIA Geforce GTX 1650 Max-Q Design
  • Memory: 16 gb

Set up Environment

Several elements need to be taken care of on order:

• Windows WSL enable & Nvidia Drivers installation
• WSL Kernel update
• Ubuntu download

Windows

• Follow the instructions to install the WSL NVIDIA DRIVER MS Youtube Video

Linux

• Setup CUDA Toolkit by following the WSL NVIDIA GUIDE MS Youtube Video

  • On step 4.2.6 (Using the WSL-Ubuntu Package) use these commands instead WSL Cuda 11.3:

  `wget https://developer.download.nvidia.com/compute/cuda/repos/wsl-ubuntu/x86_64/cuda-wsl-ubuntu.pin

  sudo mv cuda-wsl-ubuntu.pin /etc/apt/preferences.d/cuda-repository-pin-600

  wget https://developer.download.nvidia.com/compute/cuda/11.3.0/local_installers/cuda-repo-wsl-ubuntu-11-3-local_11.3.0-1_amd64.deb

  sudo dpkg -i cuda-repo-wsl-ubuntu-11-3-local_11.3.0-1_amd64.deb

  sudo apt-key add /var/cuda-repo-wsl-ubuntu-11-3-local/7fa2af80.pub

  sudo apt-get update

  sudo apt-get -y install cuda`

• Update .bashrc (located under home/) and add following lines

  `#CUDA LOCATION FROM WSL

  export LD_LIBRARY_PATH="/usr/local/cuda-11.3/lib64:$LD_LIBRARY_PATH"

  export PATH="/usr/local/cuda-11.3/bin:$PATH"

  export CUDA_HOME=/usr/local/cuda-11.3`

• run source .bashrc

• Install conda sudo wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh bash Miniconda3-latest-Linux-x86_64.sh

MMDetection conda installation

• Follow steps on the set up guide
• on step #2 use conda install pytorch=1.10.0 cudatoolkit=11.3 torchvision -c pytorch

Copy dataset

• cp -R /mnt/c/Users/johan/Downloads/val2017/val2017 ~/aigri_projects/mmdetection_sandbox/data/coco
• cp -R /mnt/c/Users/johan/Downloads/annotations_trainval2017/annotations ~/aigri_projects/mmdetection_sandbox/data/coco

Export to tensor RT

Export to ONNX

• pip install onnx onnxruntime-gpu
• Current version of mmdetection does not support onnxruntime-gpu>= 1.9.0 (But since we are using CUDA Toolkit 11.3 and that is supported on > 1.9.0) so a small tweak is needed:
  • Modify file '/home/johan/miniconda3/envs/openmmlab/lib/python3.7/site-packages/mmdet/core/export/model_wrappers.py"
    • Comment lines 115 y 123 add 'sess = ort.InferenceSession(onnx_file, session_options, providers=providers)' on line 124

Covert to TensorRT

• Donwload deb file https://developer.nvidia.com/nvidia-tensorrt-8x-download
• Follow Guide
• PIP Wheel
  • python3 -m pip install --upgrade setuptools pip
  • python3 -m pip install nvidia-pyindex
  • python3 -m pip install --upgrade nvidia-tensorrt

Using MMDET on the GPU Cluster with singularity

Loading Modules

- `module load anaconda3`
- `module load cuda-10.2`
- `module load singularity-3.6.4`
- `conda create -n openmmlabjp python=3.7 -y`

conda activate openmmlabjp

conda install pytorch==1.8.0 torchvision==0.9.0 torchaudio==0.8.0 cudatoolkit=10.2 -c pytorch

pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu102/ torch1.8.0/index.html

conda install -c anaconda ipykernel

python -m ipykernel install --user --name=openmmlabjp

sbatch conda_singularity.sh

Frequent used Commands

• sudo apt update
• sudo apt upgrade
• sudo apt install gedit -y
• stat
• sudo chown -R
• sudo apt-get --purge -y remove 'cuda*'
• nvcc --version
• wsl --update
• wsl -l -o
• wsl -l -v
• uname -a
• source .bashrc
• conda update conda
• conda env list
• conda create
• conda activate
• conda deactivate
• nvcc --version