Skip to main content


SSCMA runtime environment relies on PyTorch and various third-party libraries from OpenMMLab. You can find the SSCMA code on GitHub. To get started, make sure you have PyTorch installed locally by following the instructions here, and then obtain the necessary OpenMMLab libraries.

  • MMCV: OpenMMLab Computer Vision Foundation Library.
  • MMClassification: OpenMMLab image classification toolkit and benchmarking. In addition to classification tasks, it is also used to provide a variety of backbone networks.
  • MMDetection: OpenMMLab detection toolbox and benchmark.
  • MMPose: OpenMMLab inspection toolbox and benchmark.
  • MIM: MIM provides a unified interface for starting and installing the OpenMMLab project and its extensions, and managing the OpenMMLab model library.


SSCMA works on Linux, Windows, and macOS. We strongly recommend you to use Miniconda to manage python packages. Please follow the steps below to prepare the environment.


Miniconda is a free minimal installer for conda, you can download and install Miniconda3 from Miniconda Official Website.

Step 0 - Clone the Git Repository

First, you need to clone the SSCMA Source Code locally. We use Git to manage and host it on GitHub, and provide two different ways to clone it below (choose either one). If you don't have Git installed, you can configure Git on your computer by referring to the Git Documentation.

git clone --depth 1 && \
cd ModelAssistant

Step 1 - Create Virtual Environment

Assuming you have conda installed, then create and activate a conda virtual environment.

conda create --name sscma python=3.8 -y && \
conda activate sscma

Step 2 - Install PyTorch

SSCMA relies on PyTorch. Before running the following code, please confirm again that you have activated the virtual environment you just created.

For devices with GPUs (CUDA), we recommend installing dependencies that support GPU acceleration. We have listed the configuration options you can choose from in 2 different cases, please choose manually according to your hardware environment.

  • CPU-Only platform:
conda install pytorch torchvision torchaudio cpuonly -c pytorch
  • GPUs (CUDA) platform:
conda install pytorch torchvision torchaudio pytorch-cuda=11.7 cudatoolkit=11.7 -c pytorch -c nvidia

You can find CUDA installers on NVIDIA CUDA Toolkit Archive Website if your platform have not CUDA installed for NVIDIA GPUs, we recommend to use CUDA 11.7 or above on your host environment. For installing PyTorch on other platforms, please read more on PyTorch Official Website.

Step 3 - Install Essential Dependencies

Please confirm that you have activated the virtual environment and in the main working directory of SSCMA source code, and then run the following code to complete the configuration of the basic dependencies.

pip3 install -r requirements/base.txt && \
mim install -r requirements/mmlab.txt && \
mim install -e .

Step 4 - Install Extra Dependencies (Optional)

If you need to perform model transformation or inference testing, you also need to install the following additional dependencies.

pip3 install -r requirements/inference.txt -r requirements/export.txt

If you wish to make changes to SSCMA and submit them to us, we recommend that you additionally run the following command to facilitate checking your code at commit time.

pip3 install -r requirements/tests.txt
pre-commit install

Other Method

The configuration of SSCMA environment can be done automatically using a shell script on Linux (tested on Ubuntu 20.04~22.10), if you have Conda setup.

bash scripts/

Or you can do the configuration manually using Conda's configuration file.

conda env create -n sscma -f environment.yml && \
conda activate sscma && \
pip3 install -r requirements/inference.txt -r requirements/export.txt -r requirements/tests.txt && \
mim install -r requirements/mmlab.txt && \
mim install -e .


After completing the installation of Miniconda and configuring SSCMA with Conda, we created a Conda virtual environment named sscma and installed the dependencies in the virtual environment. For subsequent SSCMA-related configuration and development, make sure you are in the SSCMA virtual environment, which you can activate with the following command.

conda activate sscma

If you want to reconfigure or remove the SSCMA virtual environment, you can run the following command.

conda env remove -n sscma


  • I have slow connection speed while installing packages from anaconda's default channels.

    Please be patient and try some third-party mirrored channels, such as SJTU mirror, TUNA mirror and etc.

Loading Comments...