Requirements
- Python >= 3.5
- Pytorch == 1.4.0 （other versions are not tested)
- PyAV >= 6.2.0
- yacs
- OpenCV
- tensorboardX
- tqdm
- FFmpeg
- Cython, cython_bbox, SciPy, matplotlib, easydict (for running demo)
- Linux + Nvidia GPUs
We recommend to setup the environment with Anaconda, the step-by-step installation script is shown below.
conda create -n alphaction python=3.7
conda activate alphaction
# install pytorch with the same cuda version as in your environment
cuda_version=$(nvcc --version | grep -oP '(?<=release )[\d\.]*?(?=,)')
conda install pytorch=1.4.0 torchvision cudatoolkit=$cuda_version -c pytorch
# you should check manually if you successfully install pytorch here, there may be no such package for some cuda versions.
conda install av -c conda-forge
conda install cython
git clone https://github.com/MVIG-SJTU/AlphAction.git
cd AlphAction
pip install -e . # Other dependicies will be installed here