Current computer vision frameworks present APIs as lists of specific methods, which requires deep algorithmic knowledge to apply effectively in the real world. We present OpenVL, a high-level abstraction with a task-based interface which does not require extensive knowledge of or experience with vision methods. Instead we require developers to have enough knowledge of a task to accurately describe it using our interface. The description is analysed and used to invoke the appropriate algorithm and provide a solution. The underlying implementation can be optimised on combinations of CPU and GPU. The OpenVL model can describe many tasks, such as segmentation, correspondence, registration, detection, optical flow and tracking, and we will present working examples for each. We believe the adoption of the developer-friendly OpenVL interface and AMD hardware for acceleration will dramatically increase the application of computer vision technology by developers.

We will be demonstrating the latest version of OpenVL at the Summit: if you would like to see the demos or take a look at the code/interface/API, please get in touch!