Skip to content

Latest commit

 

History

History
executable file
·
38 lines (24 loc) · 1.13 KB

README.md

File metadata and controls

executable file
·
38 lines (24 loc) · 1.13 KB
Raw

Two-Stream Modeling for Egocentric Interaction Recognition

Code repository for our solution to the H2O-Action Challenge.

Install

Our code is mainly based on SlowFast, please refer to this repository for more information.

Data Preparation

You need to download the following data to start the experiment:

  1. Download the H2O dataset
  2. Extract flow frames using denseflow

Train

You can train on the H2O dataset by running:

REPO_PATH='/xxx/SlowFast'
export PYTHONPATH=$PYTHONPATH:$REPO_PATH
python tools/run_net.py --cfg configs/H2O/SLOWFAST_4x16_R50.yaml

Test

You can test on the H2O dataset by running:

REPO_PATH='/xxx/SlowFast'
export PYTHONPATH=$PYTHONPATH:$REPO_PATH
python tools/run_net.py --cfg configs/H2O/SLOWFAST_4x16_R50_TEST.yaml

Reference

The majority of this repository is borrowed from SlowFast. Thank these authors for their great work.