Installation

Quick Installation

The source code for cvtk is available on GitHub, and the built package is available on PyPI. User can install cvtk with minimal dependencies using the following command:

pip install cvtk

Installation with Full Features

For additional functionality, such as handling COCO format files, generating source code for object classification tasks with PyTorch, generating source code for object detection and instance segmentation tasks with MMDetection, and creating Flask applications, the cvtk with full dependencies should be installed.

The cvtk with full dependencies requires torch package (PyTorch) version 2.0.0 or later and mmdet package (MMDetection) version 3.0.0 or later. It is recommended that the torch and mmdet packages (and the related packages openmim, mmengine, and mmcv) be installed manually prior to installation. This is because these packages depend on the operating system (OS) and CUDA version and cannot be installed automatically. Below is an example of torch and mmdet installation in a CUDA 11.8 environment. For detailed installation instructions, please refer to the tutorial for each package.

Note

The installation of mmdet and mmcv package is prone to failure. It is important to ensure that the versions of mmdet and mmcv installed are compatible with the CUDA and torch versions. If the versions do not match, an error will occur when calling mmcv (which is called by mmdet and mmdet is called by cvtk).

# PyTorch (CUDA 11.8)
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118

# MMDetection (mim, mmengin, mmcv, mmdet)
pip install -U openmim
mim install mmengine
mim install "mmcv>=2.0.0"
mim install "mmdet>=3.0.0"

Then, user can install cvtk with all dependencies using the following command:

pip install cvtk[all]