Improvement of the detection rate of gesture outliers with a feed-forward neural network trained as the discriminator in the Generative Adversatial Network.
This is a way of solving the issue of classification of out-of-vocabulary gestures. Very often, these gestures are classified as an existing class and are difficult to remove with a classification threshold. The proposed solution has two components: (1) the use of a generative model (GAN) to augment the data set with new generated samples, (2) the use of noisy labels to decrease the average probability of predictions, thus facilitating threshold tuning.
Use the file dualmyo_gan_generator_train.py to train the generator and discriminator networks. The networks can then be tested with the scripts on tests_discriminator.py and tests_generator.py. The description of the methodology is to be published soon.
The networks are implemented in Keras. Aditional libraries used are: numpy, scikit-learn and tensorflow.
The relevant datasets are the following:
Miguel Simão:
This software is distributed under a MIT License.
Copyright (c) 2018
This work was partially funded by the Portuguese Foundation for Science and Technology (FCT) throught grant SFRH/BD/105252/2014.