by Adam Pease
This month the US Postal Service (USPS) made public news that it would be incorporating computer vision into 195 of its mail processing centers. The announcement boasted that these new AI systems would be capable of reducing the time required to track a package that has gone missing from days to hours. This announcement underscores the growing adoption of computer vision in the enterprise, and has fascinating implications for the future of infrastructure. In this blog, I review the USPS announcement and consider what it means for the future of computer vision and edge computing.
What Is the USPS Computer Vision Plan?
In 2020, the year that many critical infrastructures were challenged by the COVID-19 pandemic, the US Postal Service faced severe logistical challenges. Now, USPS has decided to overhaul its system with 7 proprietary computer vision algorithms, running on the NVIDIA EGX platform, which includes hardware accelerators for deep learning and software for scaling enterprise AI applications. The platform supports NVIDIA’s Edge Computing Infrastructure (ECIP), which the Post Office is using to manage a series of different models across a large network that has a breadth of different hardware.
While many elements of this major infrastructure investment are still being rolled out, one area where USPS will be making use of computer vision soon is in its optical character recognition (OCR) workflow. OCR recognizes text in images and converts it to editable, searchable font, an example of the kind of task that computer vision algorithms excel at completing far more quickly than legacy products. Computer vision can also enable users to automatically count similar objects in an image or video, streamlining the process of shipping boxes.
The USPS decision to work with NVIDIA for computer vision could transform the way mail is delivered.
Edge Computing and the Digital City
Edge computing is one of the vital enabling factors for scalable computer vision, and NVIDIA is one of the leading players providing solutions in this market. Its ECIP solution for deployment and underlying platform resemble a broader shift in the market towards offering distributed, edge solutions that can support a decentralized network of devices. These solutions will be especially important for markets with large, widespread underlying infrastructures or supply chains, such as shipping. Recently, Aragon published a primer on the necessary precursors to digital city infrastructure. In the cities of the future, it is likely that computer vision will power many critical systems—from security to medicine—and edge computing will be indispensable here.
In the case of the Post Office, whose algorithms are managed by Kubernetes and served through NVIDIA’s Triton Interference Server, the advantages of a flexible solution are apparent. For a federal infrastructure system that may have different levels of hardware and network enablement depending on location, it is important to find a solution that can effectively communicate with different devices. Triton delivers AI models to systems in a way that is sensitive to their GPUs, CPUs, or other capabilities. In the edge systems of the future, it will be critical to ensure that the proper algorithms are delivered to the right place, or a system may become overwhelmed or dysfunctional.
Bottom Line
The USPS decision to use NVIDIA’s architecture bodes well for the computer vision market. The market itself is picking up steam as large players realize they can experience substantial productivity gains if they are willing to do the work of overhauling their business processes. And as this story shows, the work of overhauling these processes is becoming easier as providers step in to offer solutions that are more flexible, scalable, and that have more potential for integration and edge enablement. Stay tuned, soon Aragon will be providing more detailed research about the deployment choice enterprises face when decided to roll out computer vision in the cloud or as an edge-based system.