shaipranesh2 / classification-of-body-keypoint-trajectories-of-gesture-co-occurring-with-time-expressions Goto Github PK

This project aims to develop a Neural network architecture to classify gestures with body-keypose data extracted using STGCN on ellen dataset (This code was developed as part of Google Summer of Code'23. Read more on: https://medium.com/@f20200731/google-summer-of-code-23-red-hen-labs-6bc683056ebf)

Usage:

1. ellen_npy.py is used to extract the openpose keypose points from the json format, and it also extracts the labels and creates the data using the custom dataset class. It takes in 2 arguments:- npz file data path and label path order, respectively in that order strictly. For ex: python ellen_npy.py /mnt/rds/redhen/gallina/home/ixb164/gsoc2023_ellen2014_npzs /mnt/rds/redhen/gallina/home/ixb164/labels

2. Run ellen_model.py as a python script and the model trains and validates for every epoch For ex: python ellem_model.py

The model's performance in losses and accuracy graphs are also uploaded, along with some pretrained weights and the dataset as a .pth file.