Plan based optimization of computer vision tasks
We are working on a computer vision service for our cognitive kitchen. This service should include a planning tool, that schedule and optimizes the provided algorithms. Before a plan based optimization is possible, the vision algorithms have to be reformulated as plans. This includes divide them into atomic pieces that can be reused and reorganized. A plan contains then a list of atomic pieces of code, a list of prerequisites for every part of part and an expected result. Given such a description of the vision algorithms and a set of semantically invariant transformations, our plan based optimization system can transform, optimize and parallelize those plans.
















