$ pip install torch==1.10.0+cu102 torchvision==0.11.1+cu102 -f https://download.pytorch.org/whl/torch_stable.html $ pip install mmcv-full==1.3.17 -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.10.0/index.html $ pip install mmdet==2.17.0 $ pip install icevision[all]
We currently only support Linux/MacOS installations
installation using pip
Depending on what version of cuda driver you'd like to use, you can install different versions of torch builds. If you're not sure which version to choose, we advise to use the current torch default
pip install torch==1.10.0+cu102 torchvision==0.11.1+cu102 -f https://download.pytorch.org/whl/torch_stable.html
pip install torch==1.10.0+cu111 torchvision==0.11.1+cu111 -f https://download.pytorch.org/whl/torch_stable.html
pip install torch==1.10.0+cpu torchvision==0.11.1+cpu -f https://download.pytorch.org/whl/cpu/torch_stable.html
To see what version of
cuda is installed in your current environment, run:
python -c "import torch;print(torch.__version__, torch.version.cuda)"
mmcv-fullyou can install.
mmcv-full is optional, yet it will let you unleash the full potential of
icevision and allow you to use the large library of models available in
mmdet, therefore we strongly recommend doing it.
pip install mmcv-full==1.3.17 -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.10.0/index.html pip install mmdet==2.17.0
pip install mmcv-full==1.3.17 -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.10.0/index.html pip install mmdet==2.17.0
pip install mmcv-full==1.3.17 -f https://download.openmmlab.com/mmcv/dist/cpu/torch1.10.0/index.html pip install mmdet==2.17.0
mmcv-full can be tricky as it depends on both the exact
We highly recommend that you test your installation. You can verify it by executing the following command inside your virtual environment:
curl -sSL https://raw.githubusercontent.com/open-mmlab/mmcv/master/.dev_scripts/check_installation.py | python -
If everything went fine, you should see something like the following:
Start checking the installation of mmcv-full ... CPU ops were compiled successfully. CUDA ops were compiled successfully. mmcv-full has been installed successfully. Environment information: ----------------------------------------------------------- sys.platform: linux Python: 3.8.12 (default, Oct 12 2021, 13:49:34) [GCC 7.5.0] CUDA available: True GPU 0: GeForce RTX 2060 CUDA_HOME: /usr/local/cuda NVCC: Build cuda_11.1.TC455_06.29069683_0 GCC: gcc (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0 PyTorch: 1.10.0+cu111 PyTorch compiling details: PyTorch built with: - GCC 7.3 - C++ Version: 201402 - Intel(R) Math Kernel Library Version 2020.0.0 Product Build 20191122 for Intel(R) 64 architecture applications - Intel(R) MKL-DNN v2.2.3 (Git Hash 7336ca9f055cf1bfa13efb658fe15dc9b41f0740) - OpenMP 201511 (a.k.a. OpenMP 4.5) - LAPACK is enabled (usually provided by MKL) - NNPACK is enabled - CPU capability usage: AVX2 - CUDA Runtime 11.1 - NVCC architecture flags: -gencode;arch=compute_37,code=sm_37;-gencode;arch=compute_50,code=sm_50;-gencode;arch=compute_60,code=sm_60;-gencode;arch=compute_70,code=sm_70;-gencode;arch=compute_75,code=sm_75;-gencode;arch=compute_80,code=sm_80;-gencode;arch=compute_86,code=sm_86 - CuDNN 8.0.5 - Magma 2.5.2 - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CUDA_VERSION=11.1, CUDNN_VERSION=8.0.5, CXX_COMPILER=/opt/rh/devtoolset-7/root/usr/bin/c++, CXX_FLAGS= -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp -DNDEBUG -DUSE_KINETO -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -DEDGE_PROFILER_USE_KINETO -O2 -fPIC -Wno-narrowing -Wall -Wextra -Werror=return-type -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-sign-compare -Wno-unused-parameter -Wno-unused-variable -Wno-unused-function -Wno-unused-result -Wno-unused-local-typedefs -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_VERSION=1.10.0, USE_CUDA=ON, USE_CUDNN=ON, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=ON, USE_NNPACK=ON, USE_OPENMP=ON, TorchVision: 0.11.1+cu111 OpenCV: 4.5.4 MMCV: 1.3.17 MMCV Compiler: GCC 7.3 MMCV CUDA Compiler: 11.1 -----------------------------------------------------------
Icevision is distributed in 2 different eggs:
icevision[all]- recommended - complete icevision package with all dependencies
icevision[inference]- minimal dependencies, useful for deployment or simply parsing and viewing your dataset
we recommend to install the stable release but if you want to use the most recent, bleeding edge version of the library or would like to contribute, here is how to do it:
pip install icevision[all]
pip install git+https://github.com/airctic/icevision.git@master#egg=icevision[all] --upgrade
editable mode (for developers)
git clone --depth=1 https://github.com/airctic/icevision.git cd icevision pip install -e .[dev] pre-commit install
installing using different cuda version
Installing icevision with different cuda version is possible, however it is only recommended for more experienced users.
The main constraint here is
torch versions compatibility. In short,
torch is build for a specific cuda driver version, mmcv-full on the other hand is
distributed for a specific torch build.
To see which mmcv-full wheels are available for which versions of torch, check the table at mmcv installation guide.
pip install icevision will install
icevision[inference] by default
installation using conda
The easiest way to install
icevision with all its dependencies is to use our conda
environment.yml file. Creating a conda environment is considered as a best practice
because it avoids polluting the default (base) environment, and reduces dependencies
$ curl -O https://raw.githubusercontent.com/airctic/icevision/master/environment.yml $ conda env create -f environment.yml
please note that installation may take up to 5 mins.
environment.yml works only on cuda-10.2 enabled devices. If your GPU
architecture is Ampere or newer, you have to use the pip installation method.
MMCV is not installing with cuda support
If you are installing MMCV from the wheel like described above and still are having problems with CUDA you will probably have to compile it locally. Do that by running:
pip install mmcv-full
If you encounter the following error it means you will have to install CUDA manually (the one that comes with conda installation will not do).
OSError: CUDA_HOME environment variable is not set. Please set it to your CUDA install root.
sudo apt install nvidia-cuda-toolkit
Error: Failed building wheel for pycocotools
If you encounter the following error, when installation process is building wheel for pycocotools:
unable to execute 'gcc': No such file or directory error: command 'gcc' failed with exit status 1
sudo apt install gcc
gcc (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0 Copyright (C) 2019 Free Software Foundation, Inc. This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.