/action-recognition

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

Primary LanguagePythonGNU Lesser General Public License v2.1LGPL-2.1

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
  1. 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.