Computer Vision Is a Lifecycle Problem, Not a Model Problem

About this session

Computer vision no longer fails because usable models are missing. It fails because the deployment loop around those models is fragmented: data curation, validation, architecture selection, edge deployment, and safe updates all live in different tools and under different assumptions. I’ll walk through how we approach this differently with a control layer that compresses that loop into a single system, and why optimizing time-to-learning matters more than chasing marginal accuracy gains. You’ll see real examples where recall, consistency, and failure modes matter more than benchmark scores. If you care about building vision systems that actually survive real-world conditions, this talk will change how you think about the stack.