hasaan1108/CC_ADL-Agitation_PoseC3D_v3

Changes from v2

Multiplte classes removed.

Removed Classes

• wear jacket
• take off jacket
• sneeze / cough
• touch head (headache)
• touch chest (stomachace/heart pain)
• touch back (backache)
• touch neck (neckache)
• nausea or vomiting condition
• left turn
• right turn

About the Model

Activity Recognition model for tracking activities of daily life and agitation markers using computer vision.

Dataset

Some classes were borrowed from NTU-60 Dataset while some classes were collected in-house. Classes can be viewed in the custom_labels.txt file.

Model Configuration

Tested with the following parameters

• Clip Length = 10
• Interval = 3 seconds

Useage

Compatible with mmaction2 library from openmmlab. Set up openmmlab according to the following link:

• https://mmaction2.readthedocs.io/en/latest/get_started/installation.html

1. Place the custom_skeleton.py config file in configs/skeleton/posec3d
2. Place the custom_labels.txt in tools/data/skeleton
3. Create a folder called checkpoints in mmaction2 home directory
4. Place the model_weights.pth file in the newly created checkpoints folder
5. Test out the model using demo/demo_skeleton.py file.

Testing the Model

python demo/demo_skeleton.py  ${path to the input video} ${path to the resultant output video} --config ${path to the config file (custom_skeleton.py)} --checkpoint {path to the weights (.pth file)} --label-map ${path to the label file(custom_labels.txt)}