DATASET Augmentation using GenerativeAdversarialNetworks(GANS)
The primary objective that will be focussed on in this project is increasing the size of the data set. Many techniques have been developed to combat overfitting ( plotting an exact curve along the given data which will not produce correct results during the testing phase ), these include, batch normalization, layer, etc. Nevertheless, when the data is not significant in size, the feature and parameter learning won’t proceed the way it needs to. The parameters will be underdetermined, and the network will generalize poorly. In many cases, while procuring the data, numerous numbers of samples are no available due to a various number of reasons, say, non-availability, ease-of-reach, etc. To combat the above, the data available should be put to more use by augmenting it with precise data. This can be achieved with Generative Adversarial Network model which takes care of the larger invariance space rather than restricting to the available classes.