Skip to content

Latest commit

 

History

History
284 lines (167 loc) · 7.97 KB

install.rst

File metadata and controls

284 lines (167 loc) · 7.97 KB
Raw

Installation

MMEditing is a Python toolbox based on PyTorch and MMCV. Currently, MMEditing works on Linux, Windows, macOS, and on both CPUs and GPUs. This page describes some best practices for installation.

If you have difficulties installing MMEditing, please feel free to open a discussion and describe the error as well as the steps you have tried. The community are happy to help.

Prerequisites

  • Linux (recommended) / Windows / macOS
  • Python >= 3.6
  • PyTorch >= 1.5
  • pip and conda
  • Git
  • (Only for GPU) NVIDIA GPU with driver version >= 440.33 (Linux) or >= 441.22 (Windows native)
  • (Only for macOS) Clang compiler, can be installed with xcode-select -–install

Warning

As Python 3.6 has reached end-of-life on 23-Dec-2021, we will drop support for it in the future.

Install CPU Version

MMEditing is fully supported on CPUs, despite the slow running speed. Nevertheless, the CPU version is much more lightweight and easier to configure compared to GPU versions.

For macOS, this is the only choice.

Step 1. Create and activate a conda virtual environment

conda create --name mmedit python=3.8 -y
conda activate mmedit

Step 2. Install the CPU version of PyTorch and torchvision

conda install pytorch torchvision cpuonly -c pytorch

# or "pip install torch torchvision"

If you prefer to install previous versions of PyTorch, follow this guide.

If PyTorch is already installed, check its version with:

python -c "import torch; print(torch.__version__)"
# Example output: 1.10.2

Step 3. Install pre-compiled MMCV for CPUs

pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cpu/torch1.10/index.html

Note

The link to MMCV depends on the version of PyTorch. The above link is for PyTorch 1.10.x. If you use a different version of PyTorch, you need to change the link accordingly.

E.g. https://download.openmmlab.com/mmcv/dist/cpu/ torch1.9 /index.html for PyTorch 1.9.x.

Note

Precompiled packages are available for Linux and Windows only. On macOS, install MMCV with the following command.

CC=clang CXX=clang++ CFLAGS='-stdlib=libc++' MMCV_WITH_OPS=1 pip install -e .

See MMCV documentation for more information.

A fallback option is to install the lite version of MMCV via pip install mmcv. However, it makes deformable convolution unavailable, and several models won't work.

See MMCV installation guide for more information.

Step 4. Install MMEditing

To make full utilization of configuration files and useful tools, we recommend installing MMEditing from source codes.

git clone https://github.com/open-mmlab/mmediting.git
cd mmediting
pip install -v -e .

# or "pip install -v -e .[all]" to install full dependencies and enable more features

Step 5. Verify installation

As a simple test, switch to another directory (such as /home) and import mmedit in Python. For example,

cd ~
python -c "import mmedit; print(mmedit.__version__)"
# Example output: 0.13.0

Make sure the version outputs correctly. For comprehensive unit tests, run pytest ..

Install CUDA Version

To enable the full power of MMEditing, we recommend the GPU version. The only difference lies at PyTorch and MMCV. Please pay attention to the version of and the CUDA version of PyTorch.

Note

GPU is not available for macOS.

Step 1. Create and activate a conda virtual environment

conda create --name mmedit python=3.8 -y
conda activate mmedit

Step 2. Install the GPU version of PyTorch and torchvision

conda install pytorch torchvision cudatoolkit=10.2 -c pytorch

# or "pip install torch==1.10.2+cu102 torchvision==0.11.3+cu102 -f https://download.pytorch.org/whl/cu102/torch_stable.html"

Note

To support new GPU models such as GTX 30 series, CUDA 11 is required. Install CUDA-11 based PyTorch with:

conda install pytorch torchvision cudatoolkit=11.3 -c pytorch

# or pip install torch==1.10.2+cu113 torchvision==0.11.3+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html

Please check this table for minimum driver versions required for specific CUDA versions. Usually, the latest driver works well.

If you prefer to install previous versions of PyTorch, follow this guide.

If PyTorch is already installed, check its version and CUDA version in Python:

python -c "import torch; print(torch.__version__)"
# Example output: 1.10.2
python -c "import torch; print(torch.version.cuda)"
# Example output: 10.2

Step 3. Install pre-compiled MMCV for GPUs

pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.10/index.html

Note

The link to MMCV depends on the version of PyTorch and the CUDA version of PyTorch. The above link is for PyTorch 1.10.x and CUDA 10.2. If you use a different version of PyTorch, you need to change the link accordingly.

E.g. https://download.openmmlab.com/mmcv/dist/ cu101 / torch1.8 /index.html for PyTorch 1.8.x with CUDA 10.1.

See MMCV installation guide for more information.

Step 4. Install MMEditing

To make full utilization of configuration files and useful tools, we recommend installing MMEditing from source codes.

git clone https://github.com/open-mmlab/mmediting.git
cd mmediting
pip install -v -e .

# or "pip install -v -e .[all]" to install full dependencies and enable more features

Step 5. Verify installation

As a simple test, switch to another directory (such as /home) and import mmedit in Python. For example,

cd ~
python -c "import mmedit; print(mmedit.__version__)"
# Example output: 0.13.0

Make sure the version outputs correctly. For comprehensive unit tests, run pytest ..

Another option: Install via MIM

MMEditing can be installed via MIM, a package manager dedicated to OpenMMLab projects. See MIM documentations for instructions.

Another option: Docker Image

We provide a Dockerfile for building a docker image.

To build the image:

# build an image with PyTorch 1.5, CUDA 10.1
docker build -t mmediting docker/

Run with:

docker run --gpus all --shm-size=8g -it -v ${DATA_DIR}:/mmediting/data mmediting

After installation, you can run some demos, click next.