We provide instructions to install the required dependencies.

Requirements:

• python>=3.6
• pytorch==1.5 (should work with pytorch >=1.5 as well but not tested)

1. Clone the repo:

   git clone https://github.com/TheShadow29/VidSitu.git
   cd VidSitu
   export ROOT=$(pwd)
   

2. To use the same environment you can use conda and the environment file vsitu_pyt_env.yml file provided. Please refer to Miniconda for details on installing conda.

   MINICONDA_ROOT=[to your Miniconda/Anaconda root directory]
   conda env create -f vsitu_pyt_env.yml --prefix $MINICONDA_ROOT/envs/vsitu_pyt
   conda activate vsitu_pyt
   

3. Install submodules:

   • Install Detectron2 (needed for SlowFast). If you have CUDA 10.2 and Pytorch 1.5 you can use:

   python -m pip install detectron2 -f \
   https://dl.fbaipublicfiles.com/detectron2/wheels/cu102/torch1.5/index.html
   

   Please refer to detectron2 repository for more details.

   • Slowfast:

   cd $ROOT/SlowFast
   python setup.py build develop
   

   • Fairseq:

   cd $ROOT/fairseq
   pip install -e .
   

   • cocoapi:

   cd $ROOT/cocoapi/PythonAPI
   make
   

   • coco-caption: (NOTE: You may need to install java). No additional steps are needed.

   Alternatively, you can run bash $ROOT/scripts/install_all.sh

   • coval:

   cd $ROOT/coval
   pip install -e .