A Visual Embedding for the Supervised Image based on self-attention

Deep learning has proven itself to be a successful set of models for learning useful semantic representations of data. In this paper, we explore the similar information and semantic information on vector space of images with self-attention mechanism. For this purpose we define a loss function and model architecture to obtain vector representations of image classes. Furthermore, we show that the image classes vector provide state-of-the-art performance on test set for measuring image feature and semantic similarities using metric learning. Our algorithm improves the test accuracy on Cifar10 from 87.1% to 94.125% and from 99.54% to 99.875% on MNIST compared to TripletNet [5].We also demonstrate using various datasets that our model learns a better representation, which is visualized in 2d Euclidean space.