SlowFast

PyTorch implementation of SlowFast Networks for Video Recognition (arxiv).

Preparing Kinetics Dataset

Download dataset with Kinetics downloader. For example, you could run:

cd Kinetics-downloader
python download.py data/kinetics-100-pruned_train.csv /data/kinetics-100/train/ -n 16 -t /data/kinetics-100/tmp/

Place .csv file under data directory, and rename as train.csv or val.csv. You also need classes.csv.

cp Kinetics-downloader/data/kinetics-100-pruned_train.csv /data/kinetics-100/train.csv
cp Kinetics-downloader/data/kinetics-100-classes.csv /data/kinetics-100/classes.csv

Setup

Create and start new Anaconda environment.

conda create -n slowfast python=3.9
conda activate slowfast

Install pre-requisites.

pip install -r requirements.txt

Add this repository to $PYTHONPATH.

export PYTHONPATH=/path/to/SlowFast/:$PYTHONPATH

Run

Configure, and run training.

vi SlowFast/slowfast/
python tools/train.py