Comparision of performance mass trained and fine tuned ViT with a generic CNN model using a novel combined loss function