Communications of the ACM
Elephant or Teapot? Why Occlusion Is the Next Great Frontier For Image Recognition
The next big challenge in the image recognition field will be occlusion.
Credit: thestack.com
The intensity of research in the field of image recognition has delivered significant gains in recent years, driven in large part by neural networks and advances in database classification. Scientists at Facebook AI Research (FAIR) say classic challenges such as image classification, edge detection, object detection, and semantic segmentation are so near to being solved that it is time for the field to turn to its next great challenge: occlusion.
Occlusion occurs when it is difficult to identify a given object because it is somehow obscured or obstructed in the frame, either by other objects or by tricks of perspective. In a recent paper, FAIR researchers asked human subjects to complete vector outlines for subjects in images in which they were occluded. The task involved, for example, distinguishing between a musician and an instrument they were holding in front of them or distinguishing the head of a stag from a thicket of branches. The researchers note humans were far better at identifying the occluded objects than artificial-intelligence systems.
Occlusion is expected to be a tricky issue to tackle because its most extreme manifestations yield optical illusions that frequently confound even human subjects, with one example being the Kaniza Triangle.
From The Stack (UK)
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA
No entries found