Skip to content



We currently only support Linux/MacOS installations


Please do not forget to install the other optional dependencies if you would like to use them:

  • MMCV+MMDetection, and/or

  • YOLOv5


Before proceeding with the installation, install numpy: pip install numpy

A- Installation using pip

Option 1: Installing from pypi repository [Stable Version]

To install icevision package together with almost all dependencies:

$ pip install icevision[all]

Option 2: Installing an editable package locally [For Developers]


This method is used by developers who are usually either:

  • actively contributing to icevision project by adding new features or fixing bugs, or

  • creating their own extensions, and making sure that their source code stay in sync with the icevision latest version.

Then, clone the repo and install the package:

$ git clone --depth=1
$ cd icevision
$ pip install -e .[all,dev]
$ pre-commit install

Option 3: Installing a non-editable package from GitHub:

To install the icevision package from its GitHub repo, run the command here below. This option can be used in Google Colab, for example, where you might install the icevision latest version (from the master branch)

$ pip install git+git://[all] --upgrade

B- Installation using conda

Creating a conda environment is considered as a best practice because it avoids polluting the default (base) environment, and reduces dependencies conflicts. Use the following command in order to create a conda environment called icevision

$ conda create -n icevision python=3.8 anaconda
$ conda activate icevision
$ pip install icevision[all]

Optional dependencies

MMDetection Installation

We need to provide the appropriate version of the mmcv-full package as well as the cuda and the torch versions. Here are some examples for both the CUDA and the CPU versions

Torch and CUDA version

For the torch version use torch.__version__ and replace the last number with 0. For the cuda version use: torch.version.cuda.

Example: TORCH_VERSION = torch1.7.0; CUDA_VERSION = cuda101

CUDA-Version Installation Example

$ pip install mmcv-full=="1.2.5" -f --upgrade
$ pip install mmdet

CPU-Version Installation

$ pip install mmcv-full=="1.2.5+torch.1.7.0+cpu" -f --upgrade
$ pip install mmdet

YOLOv5 Installation

$ pip install yolov5-icevision --upgrade