This demo was trained on The 20BN-jester Dataset V1 (https://20bn.com/datasets/jester/v1)
Clone the repository. You might need to install some libs such as:
- PyTorch
- OpenCV2
- Numpy
- Pandas
- Torchvision
Run demo.py
Press Q
to stop.
This is a custom NN using 2D and 3D CNN layers to achieve online video recognition. The model is trained on the Jester dataset, and adapted to the other dataset via trasnfer learning.
The data used to train and test can be downloaded in the following links:
- Jester: https://20bn.com/datasets/jester
- nvGesture:https://research.nvidia.com/publication/online-detection-and-classification-dynamic-hand-gestures-recurrent-3d-convolutional
- EgoGesture: http://www.nlpr.ia.ac.cn/iva/yfzhang/datasets/egogesture.html
- Example of trainning
- Add pre-trained models for EgoGesture and nvGesture
Special thanks to @ahmetgunduz and @ms3001 for providing some of the code used here.
Please cite the following article if you use this code or pre-trained models:
@ARTICLE{9085286,
author={F. {Brandolt Baldissera} and F. L. {Vargas}},
journal={IEEE Latin America Transactions},
title={A Light Implementation of a 3D Convolutional Network for Online Gesture Recognition},
year={2020},
volume={18},
number={02},
pages={319-326},
doi={10.1109/TLA.2020.9085286}}