Action Recognition

Three steps to train your own model for action recognition based on CNN and LSTM.

Environment

Python 3.7.5 (Anaconda 5.3.0):

https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/

PyTorch 1.3.0 (CUDA 10.1):

https://pytorch.org/

Data

1. Download the dataset (HMDB51):

python data/download.py

2. Convert videos to images:

python data/video2jpg.py

Train

Use the following command to train the model:

python train.py

Network

This project is based on CNNs and LSTMs.

Tips

If you download the dataset by yourself, you need to move the rar file to data folder firstly.