Génération de données Synthétiques pour les Datasets

Abstract

The image augmentation algorithm was used to increase the number of certain images due to the unbalanced dataset and to increase the robustness of the model. In this project, CycleGAN, one of the conditional GAN models, was used to augment image data for disease detection. We have proposed to use a variant that allows image-to-image translation without the need for paired examples of transformation from source domain to target domain. It can transform the image from one domain to another without one-to-one mapping between source domain and target domain. The tests carried out on the Plantvillage dataset gave very satisfactory results.