[NEW!] 2022 Ego4D Challenges now open for Forecasting
- Short Term Anticipation (deadline June 1 2022)
- Long Term Anticipation (deadline June 1 2022)
- Future Hand Prediction (deadline Oct 1 2022)
This repository contains code to replicate the results of the EGO4D Forecasting Benchmark in Ego4D: Around the World in 3,000 Hours of Egocentric Video.
EGO4D is the world's largest egocentric (first person) video ML dataset and benchmark suite.
For more information on Ego4D or to download the dataset, read: Start Here.
This code requires Python>=3.7 (this a requirement of pytorch video). If you are using Anaconda, you can create a clean virtual environment with the required Python version with the following command:
conda create -n ego4d_forecasting python=3.7
To proceed with the installation, you should then activate the virtual environment with the following command:
conda activate ego4d_forecasting
Run the following commands to install the requirements:
cat requirements.txt | xargs -n 1 -L 1 pip install
In order to make the ego4d module loadable, you should add the current directory to the Python path:
export PYTHONPATH=$PWD:$PYTHONPATH
Please note that the command above is not persistent and hence you should run it every time you open a new shell.
Please refer to the following README files for the benchmark specific code/instructions:
python -m ego4d.cli.cli
--output_directory=.
--datasets lta_models
--benchmarks FHO