Automatic Synthesis of Computer Vision Workflow using AI Planning

The creation of a workflow for solving computer vision problems is a complex task. The current practice largely rely on domain experts to achieve this. The search space for creating a suitable solution using available algorithms for a given goal is large. This exploratory work of solution building is time-, effort-and intellect-intensive endeavor. To address these issues, we propose a structured and generalized goal-driven algorithm selection approach for building computer vision workflows on the fly. It generates workflows depending on initial conditions and goal conditions by combining various image processing algorithms. Symbolic AI planning is aided by Reinforcement Learning to recommend optimal workflows that are robust and adaptive to changes in the environment. Experimental results show that our proposed framework gives significantly better workflows as compared to the template-based systems.