Google presented a good number of papers for computer vision research and one of them teaches computers to recognize and describe relevant objects in an image. Image INQUIRER.net
With the Computer Vision and Pattern Recognition conference in Las Vegas on the way, Google has announced several of its accomplishments.
The company’s researchers have, among others, successfully taught computer vision systems to detect specific objects in an image. Tech Crunch reports that Google’s system, developed in collaboration with Stanford, is able to track objects and determine which one is the most relevant. It can also point out objects with potential relevance as a scenario progresses.
Another accomplishment is a system that detects objects with continuously moving parts or “articulated object classes.” By identifying moving parts and position relative to the greater whole, limbs can be highlighted frame by frame. Such a system could greatly benefit surveillance systems if fully developed.
Finally, the company has developed a method for computer vision systems to not only identify different objects in an image but also describe them in detail.
The computer will be able to look through a number of provided descriptors while analyzing a picture and apply said descriptors or a combination of them onto certain objects that it detects, such as a gray laptop or a red and black watch on a white table.
While we humans can do this instinctively with our collection of life experiences and memories, computers still find this action quite difficult.
All the systems mentioned use the deep learning system. Alfred Bayle