manolisfrag / signing_or_not_classifier

Signing or not Classifier

Classifying whether a person is signing or not.

Pose estimation is highly based on alesolano's code here: https://gist.github.com/alesolano/b073d8ec9603246f766f9f15d002f4f4 Read also his article on: https://arvrjourney.com/human-pose-estimation-using-openpose-with-tensorflow-part-2-e78ab9104fc8 His code is also based on Ildoo Kim's code (https://github.com/ildoonet/tf-openpose) and derived from the OpenPose Library (https://github.com/CMU-Perceptual-Computing-Lab/openpose/blob/master/LICENSE)

Getting Started

Fork the project
$ mkdir models
$ cd models
$ wget https://www.dropbox.com/s/2dw1oz9l9hi9avg/optimized_openpose.pb
$ cd ..
$ python main.py --video 'path/to/your/video'

Alternatively:

To run the model with XGBoost run Model.ipynb notebook

Prerequisites

Tensorflow
Keras
Pympi
Pillow
XGBoost

Classifiers

Four differenent classifiers have been tested. The main.py uses the neural network created in /Classifiers/ Keras_classifier. The Model.ipynb uses the XGBoost model created in /Classifiers/SVM_RF_XGBooster.ipynb

Datasets

The datasets that have been used for the training are located in the Classifiers folder

Acknowledgments

• [alesano] (https://gist.github.com/alesolano)
• [Ildoo Kim] https://github.com/ildoonet/tf-openpose)
• [CMU-Perceptual-Computing-Lab] (https://github.com/CMU-Perceptual-Computing-Lab/openpose)
• [dopefishh] (https://github.com/dopefishh/pympi)