LightAction, a finegrained Action Recogntion Built on Mediapipe

output_feed.mp4

Step 1 : Install Pycharm

Head over Pycharm Website and Install Pycharm. I use the Community or Edu edition.
Create a new Pycharm project
for convienience I am using Python 3.8+

Clone repository and Install requirements

Open Pycharm Terminal and type
create a new project
clone this repository

git clone https://github.com/exponentialR/lightaction.git

Install requirements

run requirements.txt

Prepare Data Folder

Open Pycharm terminal and go into the cloned repository

cd data

Below you can define the following the type of actions you are collecting data for. In this case I am collecting data for OK NO WAVING CLAPPING and SALUTE The Subfolder argument takes an integer and automatically creates 30 subfolders for each actions. The --pat argument is the parent name of the data folder. The --video_length takes an integer value. It is the length of the seqeunce each actions will be running for.

 C:\PycharmProjects\iamsMediapipe\data> python create_dir.py --actions OK NO WAVING CLAPPING SALUT
E --subfolder_numbers 30 --pat actionData --video_length 60

If you have everything set up correctly you should have the followingas depicted in the screenshot.

Drawing only plots/landmarks

To plot only the keypoints at realtime. Edit the run.py for the following, or create a new python script by calling the follwowing

Plotting only the hands data.draw_only_hands
Plotting only the face data._only_face
plotting only the pose data._only_pose
plotting all hand, pose, facial data._whole_body

Data Collection

There are 4 Options for data collection as provided in the code. Edit run.py by uncommenting the following in accordance to what you are trying to do.

Collecting hand data only #data.collect_data_HAND
Collecting face data only #data.collect_data_FACE
Collecting pose data only #data.collect_data_POSE
Collecting all hand, pose, facial data only #data.collect_data_WHOLE

Train

To train your action recognition check out IAMSaction.ipnyb