How to use it

1. Install additional environment for MCG-NJU/MGSampler:

pip install -U scikit-image
pip install ffmpeg-python

2. Download the loading.py provided by this project.
3. Clone original mmaction2 project.

git clone https://github.com/open-mmlab/mmaction2

4. Aftering cloning, in original mmaction2 folder, replace mmaction2/mmaction/datasets/pipelines/loading.py to the loading.py provided by this project.

rm -rf mmaction2/mmaction/datasets/pipelines/loading.py
cp loading.py mmaction2/mmaction/datasets/pipelines/loading.py

5. Go to the mmaction2 folder and install mmaction2 folder with new loading.py.
   If you install original mmaction2 before, please remove the old version and then re-install this new version.

cd mmaction2
pip install -r requirements/build.txt
pip install -v -e .  # or "python setup.py develop"

Overview

This repo is just use Image-level, not Fature-level This is forked from MCG-NJU/MGSampler. Combined its part of changes for easily matching to new version mmaction2 in future use. The git project MCG-NJU/MGSampler is the implementation of MGSampler and its code is based on mmaction2.