Relative Attributes for Enhanced Human-Machine Communication

We propose to model relative attributes1 that capture the relationships between images and objects in terms of human-nameable visual properties. For example, the models can capture that animal A is ‘furrier ’ than an-imal B, or image X is ‘brighter ’ than image B. Given training data stating how object/scene categories re-late according to different attributes, we learn a rank-ing function per attribute. The learned ranking func-tions predict the relative strength of each property in novel images. We show how these relative attribute pre-dictions enable a variety of novel applications, includ-ing zero-shot learning from relative comparisons, auto-matic image description, image search with interactive feedback, and active learning of discriminative classi-fiers. We overview results demonstrating these applica-tions with images of faces and natural scenes. Overall, we find that relative attributes enhance the precision of communication between humans and computer vision algorithms, providing the richer language needed to flu-idly “teach ” a system about visual concepts