This framework focuses on Human Activity Recognition (HAR) problems by using machine learning techniques, fine-tuning based on pretrain model such as ResNet50 and Bi-LSTM in combination. The goal is to create a model that can classify human activities when user feed model videos (in any terms of camera angles, resolution, quality). <--video demo shortly-->

Note: CSC16004 – Computer Vision - Final Project Rework - 22TGMT HCMUS

• 3-12-2025: Add Hugging Face Space for Web Inference demo.

git clone https://github.com/BAoD1nH/BD_HAR_25.git --recursive
cd BD_HAR_25

pip install -r Source/requirements.txt

python Source/dataset_download.py

python Source/main.py <type-of-dataset> <dataset-path>

E.g.

python Source/main.py ucf11 dataset/ucf11_updated_mpg

python Source/Inference.py <type-of-dataset> <test_video-path> <model-path>

E.g.

python Source/inference.py ucf11 test_videos/basketball.mp4 Source/models/ucf11_lstm_model.pt

• [Hoàng Bảo Khanh] - [github/hbkhanh22], FIT HCMUS-VNU
• [Đinh Nguyễn Gia Bảo] - [github/BAoD1nH], FIT HCMUS-VNU

• This project is licensed under the MIT License - see the LICENSE file for details.
• Any questions please contact via [email protected]

If you find this project useful, please give it a star ⭐️! Contributions are also welcome.