The main goal of this work is to develop system that will be able to detect any action asked.
For the first part of the project, I aim for implementing a pioneering study of tube convolutional networks. To achieve this, I use Python and PyTorch framework for CNN development.
To test the model UCF-50 Sports Action Detection (https://www.crcv.ucf.edu/data/UCF_Sports_Action.php) dataset was used, which contains 10 types of actions.
Updates concerning part 2 will arrive soon.
- Clone this repo to your local machine using
https://github.com/cyberpunk317/Action_detection.git
- Firstly, install the required packages
$ pip install -r requirements.txt- Train model from scratch with the following command
$ python training.py- Evaluate model on your video
(currently in development)