Implication Of Manifold Assumption In Deep Learning Models For Computer Vision Applications