Deploying Computer Vision by Combining Deep Learning Action Sets with Open-Source Technology

Though early computer vision dates back as far as 1927, it has gained momentum in recent years due to the developments in the fields of deep learning and artificial intelligence. With the desire for applying computer vision in ever more general and flexible contexts, the challenge arises how we can push image processing models to production in a robust way. This paper focuses on doing Automated Meter Reading (AMR) using customer-submitted photos of their meters. The challenges in this context are two-fold.