From Visual Recognition to Visual Understanding
Date:
Time:
19:00
Location:
Aula
Advances in deep learning have led to tremendous progress in systems for visual recognition. As opposed to ten years ago, visual-recognition systems are now good enough that they can be used in a range of practical applications.
The talk gives an overview of how these visual-recognition systems work, and presents our recent work that improves the quality of these systems even further by training them on billions of web images. Finally, the talk discusses the difficulty of building systems that go beyond recognition and develop visual understanding capabilities. Specifically, it presents new benchmark tasks that are designed to better understand the limitations of current systems for visual understanding.