Visipedia circa 2015

• Survey of research pertaining to Visipedia project.
• Account of real-world system for bird species identification.
• Framework for characterizing schools of thought in crowdsourcing based binary annotation.
• Methods for crowdclustering and perceptual embedding estimation for category discovery and attribute induction.
• Discussion of open problems and research challenges involving systems composed of humans and machines.

Visipedia is a network of people and machines designed to harvest and organize visual information and make it accessible to anyone who has a visual query. We discuss technical challenges arising from Visipedia and discuss their implications for pattern recognition, computer vision, machine learning and visual psychology. Amongst these are discovering visual information that is implicit in experts’ brains and in crowds of people and estimating its accuracy. To motivate our thinking we explore a case study, an automated field guide to the birds of North America. We conclude by discussing research directions that are necessary to make progress on Visipedia. An important realisation is that the study of ‘computer vision’ and ‘machine learning’ has to be broadened to include the process of information discovery and the dynamic interaction of people and machines in this context. Human-machine systems with no oracle are now within the scope of pattern recognition, machine learning and computer vision.

Figure optionsDownload high-quality image (207 K)Download as PowerPoint slide