This repository contains a set of four CNNs (Convolutional Neural Networks) which can be trained to recognize actions in still images and videos based on optical flow.
- Download the image and video datasets and add them to the data folder
- Run main.py -> this will preprocess the data as well as train the CNNs
- Do your predictions!
[Note]: Every time you run main.py after the first time, it will recognize the data and CNNs that are already trained (and saved), and won't train them again.
Install the following prerequisites:
Packages:
- tensorflow == "1.8"
- matplotlib
- numpy
- h5py
- pandas
- seaborn
- opencv-python
- scikit-learn == "0.22.2"
Requires:
- python_version = "3.6" or higher
- Lorenzo Bini
- Neele Dijkstra