siddhantkumar14 / dynamic-webcam-gesture-recognition

This is our work on the open source data 20bn Jester, which contains about 25 different hand gestures performed in front of a webcam. We will be using Optical Flow and 3 dimensional CNNs primarily with other computer vision techniques.

Dynamic-Webcam-Gesture-Recognition

This is my work on the open source dataset 20bn Jester, which contains 25 different hand gestures performed in front of a webcam. We will be using 3 dimensional CNNs primarily with other computer vision techniques.

I have implemented the model based on this paper, which has a good response time while maintaining the accuracy of the model. The model I have trained has a testing accuracy of 93.7% at 18 epochs (weights - main.h5), which is where it plateaus. I have also written a custom VideoDataGenerator that works in a similar fashion to keras' ImageDataGenerator for 3D neural networks, do use it :)

Run predict.py with the necessary libraries (keras, tf, opencv, matplotlib, numpy and os) installed -

OpenCV - 4.4.0

Tensorflow - 2.0.0

Keras - 2.1.1

OR

Run the command - 'conda env create -f 3Dgesture.yml' to automatically create the environment with the necessary libraries

dynamic-webcam-gesture-recognition's Issues

import vlc，The library could not be found

the output of cell " np.shape(inst[0]) " only (2,)