Human Activity Classification - is our project for High-Level Computer Vision for Summer Semester 2022.
In this project, we use pre-trained models and fine tune them to classify human activity. We use a publicly available image dataset found from Kaggle. The dataset is comprised of 15 different classes. We purposely used a small dataset to address the overfitting issues with respect to other proposed models and tried to mitigate it using a deep ensemble technique. Our ensemble method uses a weighted sum model to achieve best performance.
https://www.kaggle.com/datasets/meetnagadia/human-action-recognition-har-dataset
We used hardware support provided by Kaggle. Kaggle is a free website that allows users to train models online and participate in machine learning competitions.
Shantanu Kumar Rahut (Matriculation No. : 7015438)
Rushan Mukherjee (Matriculation No. : 7015520)