INSTALL.md

Installation

Requirements

• Python >= 3.7
• Pytorch = 1.7.1
• yacs
• OpenCV
• tensorboardX
• tqdm
• FFmpeg
• Linux and Nvidia GPUs (we used one 1080 GPU for J-HMDB and UCF24, 8 Tesla v100 GPUs for AVA and MultiSports)

We recommend to setup the environment with Anaconda, the step-by-step installation script is shown below.

conda create -n hit python=3.7
conda activate hit

# install pytorch with the same cuda version as in your environment
cuda_version=$(nvcc --version | grep -oP '(?<=release )[\d\.]*?(?=,)')
conda install pytorch torchvision cudatoolkit=$cuda_version -c pytorch

conda install av -c conda-forge
conda install cython
pip install einops

git clone https://github.com/joslefaure/HIT.git
cd HIT
pip install -e .    # install other dependencies

Errors frequently encountered + solutions

1. If the last command fails with No CUDA found error, it might be because Anaconda installed the CPU version of pytorch. Please use pip install with your prefered pytorch and cuda versions and try again. For instance, with torch 1.10.0 and cuda 10.2, you would run: pip install torch==1.10.0+cu102 torchvision --extra-index-url https://download.pytorch.org/whl/cu102

2. If the error is fatal error: THC/THC.h: No such file or directory, I suggest keep your pytorch version between 1.4.0 and 1.10.0, inclusive.

3. If you use a more recent GPU such as a 3080, please try these commands to install hit:

TORCH_CUDA_ARCH_LIST="8.0" python setup.py install
export TORCH_CUDA_ARCH_LIST=8.0

4. Any other issues? Please open an issue.