Cats and dogs are among the many objects people are pretty good at recognizing, but computers are not. “Look, this is a cat!” and “Look, that’s a dog!” are cries you are more likely to hear from a person than from a silicon-based computer. (In truth, you are not all that likely to hear people shout those exact statements — but you are very unlikely to hear computers spontaneously ejaculate them.)
But many people, like most computers, aren’t all that great at recognizing which kinds of cats and which kinds of dogs they see.
Computer scientists at the University of Oxford in England and the International Institute of Information Technology, in Hyderabad, India, are working on making computers better at distinguishing a cat from a dog from anything that is neither. They are also working to reliably, automatically recognize what species of dog the dog is, and what kind of cat the cat.
The biggest problem is that cats and dogs come in many shapes. This shape-shiftiness intrigues computer scientists. The Oxford/Hyderabad team explains that researchers “have often focused on cats and dogs as examples of highly deformable objects for which recognition and detection is particularly challenging.”
The team produced two especially cat-and-dog-centric studies about all this. The images you see here are from those studies. The second study focuses on the “which species is this?” question:
“Beyond the technical interest of fine grained categorization, extracting information from images of pets has a practical side too. People devote a lot of attention to their domestic animals, as suggested by the large number of social networks dedicated to the sharing of images of cats and dogs: Pet Finder, Catster, Dogster, My Cat Space, My Dog Space , The International Cat Association and several others. In fact, the bulk of the data used in this paper has been extracted from annotated images that users of these social sites post daily. It is not unusual for owners to believe (and post) the incorrect breed for their pet, so having a method of automated classification could provide a gentle way of alerting them to such errors.”
The studies are:
“The truth about cats and dogs,” Omkar M. Parkhi, Andrea Vedaldi, C. V. Jawahar, and Andrew Zisserman, in Computer Vision (ICCV), 2011 IEEE International Conference on, pp. 1427-1434. IEEE, 2011.
“Cats and dogs,” Omkar M. Parkhi, Andrea Vedaldi, Andrew Zisserman, and C. V. Jawahar, in Computer Vision and Pattern Recognition (CVPR), 2012 IEEE Conference on, pp. 3498-3505. IEEE, 2012.
UPDATE (July 18, 2014, 5 pm): A reverse twist on this: Asirra (Animal Species Image Recognition for Restricting Access) — a system that serves the role of asking for a password, but does it more pleasantly by instead “asking users to identify photographs of cats and dogs”. (Thanks to Florian Gallwitz for bringing it to my attention.) Asirra is the work of three Microsoft Research researchers: John (JD) Douceur, Jeremy Elson, and Jon Howell. [Note: I should, and do, disclose that Microsoft Research is a financial supporter of this year’s Ig Nobel Prize Ceremony.] Here, in contrast, is a video from the 1930s showing a more traditional way to ask for a password: