Visual attributes are great means of describing images or scenes, in a way both humans and computers understand. In order to establish a correspondence between images and to be able to compare the strength of each property between images, relative attributes were introduced. However, since their introduction, hand-crafted and engineered features were used to learn complex models for the problem of relative attributes. This limits the applicability of those methods for more realistic cases. We introduce a two part deep learning architecture for the task of relative attribute prediction. A convolutional neural network (ConvNet) architecture is adopted to learn the features with addition of an additional layer (ranking layer) that learns to rank the images based on these features. Also an appropriate ranking loss is adapted to train the whole network in an end-to-end fashion. Our proposed method outperforms the baseline and state-of-the-art methods in relative attribute prediction on various datasets. Our qualitative results also show that the network is able to learn effective features for the task. Furthermore, we use our trained models to visualize saliency maps for each attribute.
View on arXiv