Past Recording
ShareSave
Some Salient Issues with Saliency Models
Tuesday Oct 13 2020 16:00 GMT
Please to join the live chat.
Some Salient Issues with Saliency Models
Why This Is Interesting

Deep learning has come to dominate many areas of artificial intelligence. Given sufficient training data, deep learning provides unparalleled pattern matching over even the extremely high dimensional data involved in visual processing, and thus deep learning has become the basis for many state-of-the-art approaches in the field of computer vision. Despite this widespread success, however, it is important to consider whether deep learning approaches are always solving the tasks we would like them to be. Using the domain of visual saliency modeling as an example, I will highlight two major challenges facing deep learning applications in computer vision: data-fitting alone is not always sufficient, and the need to be able to relate enclosed deep learning models to broader models of vision.

Discussion Points
  • What is saliency modeling?
  • How do current state-of-the-art deep learning approaches model saliency?
  • Are these models actually modeling saliency, and what are some modes of failure?
  • If we want to use these models to understand human vision, what is missing?
Time of Recording: Tuesday Oct 13 2020 16:00 GMT