MMDetection3D Library Installation Guide

calendar Oct 30, 2023 · 2 min read · AI  ·
Share on: linkedin copy

Overview

Cuda, Conda, MMDetection3D installation and test.

1. Prepare

The installation is tested in environment: Ubuntu 20.04, CUDA-11.6, PyTorch v1.13.1.

1.1 Miniconda

The installation is inside a conda virtual envrionment, so be sure Miniconda is installed.

Installation on Linux:

1mkdir -p ~/miniconda3
2wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O ~/miniconda3/miniconda.sh
3bash ~/miniconda3/miniconda.sh -b -u -p ~/miniconda3
4rm -rf ~/miniconda3/miniconda.sh

And initialization:

1~/miniconda3/bin/conda init bash
2~/miniconda3/bin/conda init zsh

1.2 Remove and Install

CUDA 11.6 and Nvidia driver 510.39.01 are recommended in the guide. The Nvidia driver can be installed along with CUDA.

Remove Cuda

  • Check cuda version: nvcc --version
  • Remove cuda: 
    1sudo apt-get --purge remove "*cuda*" "*cublas*" "*cufft*" "*cufile*" "*curand*" "*cusolver*" "*cusparse*" "*gds-tools*" "*npp*" "*nvjpeg*" "nsight*" "*nvvm*"
  • or sudo /usr/local/cuda-11.x/bin/cuda-uninstaller (replace x with the version of cuda)

Remove Nvidia Driver

  • Check nvidia driver version: nvidia-smi
  • Remove nvidia driver: sudo /usr/bin/nvidia-uninstall or sudo apt-get --purge remove "*nvidia*" "libxnvctrl*"
  • sudo apt-get autoremove (caution)

Installation

Option 1

Install CUDA 11.6 and driver 510.39.01:

1wget https://developer.download.nvidia.com/compute/cuda/11.6.0/local_installers/cuda_11.6.0_510.39.01_linux.run
2sudo sh cuda_11.6.0_510.39.01_linux.run

Select the Nvidia driver during CUDA installation in Terminal.

For other versions, check method from Nvidia toolkit website.

Option 2

Open Software & Updates, click Additional Drivers, selecet Nvidia driver and click Apply Changes.

2. Install dependencies

2.1 Create envrionment from file

Download the envrionment.yml file. Create the envrionment by:

1conda env create --file environment.yml

Once created successfully, jump to Section 3.3 to test it.

If otherwise failed, follow Section 2.2 to install it manually.

2.2 Manual installation

Create a virtual environment:

1conda create --name openmmlab python=3.8 -y
2conda activate openmmlab

Install Pytorch v1.13.1:

1conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.6 -c pytorch -c nvidia

Install SpConv:

1pip install spconv-cu116

3. Install MMDetection3D

See more from: https://mmdetection3d.readthedocs.io/en/latest/get_started.html

3.1 Install MM* dependencies

1# Install mim
2pip install -U openmim
3mim install mmengine 'mmcv>=2.0.0rc4,<=2.2.0' 'mmdet>=3.0.0'

3.2 Install MMDetection3D library

Case one: use mmdet3d as a dependency or third-party package:

1# Install mmdet3d (e.g. 1.4.0)
2mim install "mmdet3d>=1.1.0"

Case two: develop and run mmdet3d directly, install it from source:

1git clone https://github.com/open-mmlab/mmdetection3d.git -b dev-1.x
2cd mmdetection3d
3pip install -v -e .

After installation, verify the package location by pip show mmdet3d.

3.3 Test

To test the result, try to import libraries from it in a Python IDE, for example:

1from mmdet3d.apis import init_model, inference_detector

or

1import mmdet3d
2print(mmdet3d.__version__)
comments powered by Disqus