Build Computer Vision Solutions Faster with Intel Geti and Modzy

Computer vision can now power seemingly endless functions across locations like manufacturing plants, retails stores, and secure facilities. From improving quality assurance processes in manufacturing, to enhancing inventory tracking in retail, to monitoring worker safety in potentially hazardous working conditions, opportunities abound. But the process of building and rolling-out these functions is still quite difficult, due to the struggle of training custom computer vision models, the difficulty of running computer vision-powered applications across a mix of hardware, or the hassle of managing the entire lifecycle.

Fortunately, Intel and Modzy have partnered together to offer a solution that combines the power ofIntel’s Geti platformfor computer vision model training with Modzy’s platform for deploying, connecting, and running AI models anywhere. Intel Geti provides customers an interactive, no-code model training platform that radically simplifies the process of training a new model from just a handful of images. These models can then be imported into Modzy, aschassis.mlmodel containers, for deploying, running, and monitoring them anywhere in the world– in the cloud, on-prem, or at the edge. Best of all, these models are optimized withOpenVINO, amping up their performance across a wide range of Intel and ARM devices.

The shortened time-to-value that this solution delivers is twofold: it provides the ability for less technical end users to quickly develop powerful machine learning models, and it allows them to easily integrate computer vision capabilities into powerful production applications.

何w it Works

The Intel Geti x Modzy integrated solution combines two powerful platforms to solve some of the biggest hurdles to developing computer vision solutions today – building computer vision models from scratch, and moving models out of the lab into scaled production applications.

The Intel Geti platform democratizes access to AI by providing tools for teams to build high quality computer vision models. It facilitates collaboration between data scientists, ML experts, and domain experts to quickly and easily develop computer vision models.

Once models are trained, tested, and validated in Intel Geti, they can be exported into the OpenVINO format. The only additional step needed before a model can be deployed to Modzy, is to package it into an Open Model Interface (OMI) model container. OMI containers fully support the OpenVINO IR format on either the ONNX or OpenVINO runtime. This makes it easy to create an optimized model container that will always run performantly on Intel CPUs and GPUs, and even some ARM devices. Later this summer, chassis.ml will be launching a new way to automatically create OpenVINO optimized model containers which will make the model export process totally seamless.

From there, Modzy deploys these models into production and turns them into API endpoints that can be integrated into applications running anywhere at scale, including on edge devices such as Raspberry Pis, NVIDIA Jetsons, and Intel NUCs. Many computer vision applications require real-time inference and insights, meaning they must be able to run at the edge, so this piece is critical.

Applications by Industry

This integrated solution can be used to develop computer vision solutions at the edge for manufacturing, retail, construction, energy & utilities, and more.

For manufacturers, this solution could be used to develop visual inspection and defect detection solutions that identify, classify, and report defects in production line processes. It also could be used to develop solutions that conduct anonymized video safety monitoring, powering applications that monitor worker safety for things like PPE detection, or unsafe behavior while blurring employee faces to ensure anonymized reporting.

In retail, the integrated Intel Geti x Modzy solution could be used to develop computer vision solutions for inventory tracking to count and report stock levels of various items. It could also be used to monitor wait and queue times to improve customer experience, or to detect suspicious activity in surveillance videos.

Because of the horizontal nature of this solution, it can be used to develop any kind of computer vision application for any industry. To get started, contact[email protected].