Introducing Deepomatic Run to deploy and industrialize local AI applications
Today we are thrilled to announce the release of Deepomatic Run®, our product that lets you deploy and put your neural networks into production on local servers.
This new product adds the ability to run neural networks locally in addition to the cloud. You can still use and integrate our web APIs into your applications and experiment on your data in the cloud. Now, you can also rely on Deepomatic Run® to industrialize large-scale video recognition projects on your infrastructure, this is a crucial new step.
How does it work?
We’ve been working on local server deployment for quite some time now, but we needed to package it into a complete product to make it very easy for you to deploy and update your neural networks locally.
There are a few hardware and software requirements for your server to deploy Deepomatic Run®: Standard Linux distribution with dedicated NVIDIA GPUs.
However, we are adding more and more deployment options as we speak so let us know if you need another configuration for your deployment.
Your application is composed of several orchestrated docker containers. Once set up, you’ll be able to easily configure and connect your server to a camera stream and analyse either images or videos.
Use in Production
Up until now, you’ve most likely been in contact of our main product: Deepomatic Studio®. We have designed Deepomatic Run® to walk hand in hand with Deepomatic Studio®. Each local Deepomatic Run® deployment is registered in Deepomatic Studio®. You are thus able to select which apps you want to run, and where.
Your apps both have an online and offline mode to best fit your environment. If applicable, you will be able to setup a feedback loop from your local server. More specifically, you have the possibility to automatically filter all images going through your apps and send them to Deepomatic Studio® to increase the size of your datasets with production data. You can then train new versions and deploy those new versions seamlessly on your servers. This way your algorithms keep getting better and better and always stay relevant and up-to-date.
Both products are then complementary and their conjoint use allows you to manage the entire lifecycle of your video recognition applications: from managing data and training deep learning neural networks to deploying them at industrial scale, and analyzing local video streams.
Deepomatic Run® has been designed from the root to answer the operational constraints you will encounter when deploying deep learning models locally and we can’t wait to see what you will achieve with it!
Discover more about the product here: