Dataset (Outlier & GAN)

Résumé fourni par la source

The Ekush dataset has been applied in our work which is publicly available at https://shahariarrabby.github.io/ekush/. There are two proposed methods- DCGAN(Deep Convolutional Generative Adversarial Network ) and Outlier. The outlier removed classes dataset has been given as named OnlyOutlierDataset.zip. For DCGAN, five classes have been used as there contains more imbalanced dataset. After performing DCGAN, the outlier is also applied to the corresponding classes to remove noises. DCGAN_dataset.zip contains the datasets before applying noise removed on DCGAN and after the noise removed on DCGAN. Three datasets have been classified in this experiment using the ResNet-50 classifier. First, the original dataset has been used in the classifier. Then the outlier removed classes created in OnlyOutlierDataset.zip with the rest of the original dataset has used. Finally, together with DCGAN generate images and outlier removed classes created in gan_with_outlier_and_normal.zip with the other classes of the dataset used in the classifier. These three approaches datasets are split into three sets - train, validation, and test.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.