Plainsight has launched OpenFilter, an open source project for developing, deploying, and scaling production-grade computer vision applications. The launch took place at the Embedded Vision Summit, the conference for innovators incorporating computer vision and AI in products, taking place in Santa Clara, California.

Plainsight is best known for its automated infrastructure for data-centric AI pipelines. The new tool, which is available under the Apache 2.0 license, provides a "filter" abstraction that combines code and AI models into modular components that allow developers to assemble vision pipelines.

OpenFilter is designed to simplify communication between components (called filters) and support synchronization, side-channel paths, metrics, and load balancing. It is written in Python and includes pluggable filter components, and the ability to develop and test filters locally with Python. The pipelines are designed for high-throughput synchronization, and the tool supports MQTT/REST visualization and data publishing. It also supports parallel processing via load-balanced filter branches, and the developers say built-in telemetry and metrics streaming are coming soon. 

Plainsight will demonstrate OpenFilter at the Embedded Vision Summit, where CEO Kit Merker is giving a talk titled "Beyond the Demo: Turning Computer Vision Prototypes into Scalable, Cost-Effective Solutions." on Thursday, May 22. 

OpenFilter aims to overcome the challenges enterprises face when working to deploy AI computer vision in production. The developers say OpenFilter's frame deduplication and priority scheduling reduce GPU inference costs and its advanced abstractions reduce deployment timelines from weeks to days. The team says the tool makes it easy to adapt it to audio, text, and multimodal AI. 

OpenFilter is designed to provide integrated tooling. It comes with pre-built filters for common tasks such as tracking, cropping, and segmentation. Its filter runtime manages video inputs including RTSP, webcams and image files; and takes care of  processing and output routing to databases, MQTT, or APIs.

The tool can be used to create modular reusable workflow pipelines made up of assemblies of filters for tasks including object detection, deduplication, and alerts. Users can deploy filters across CPUs, GPUs, or edge devices to optimize resource costs, and the tool comes with support for models including PyTorch, OpenCV, or custom models such as YOLO. 

Andrew Smith, CTO of Plainsight, said that as filters are the building blocks for operationalizing vision AI, the new tool means that:

"Instead of wrestling with brittle pipelines and bespoke infrastructure, developers can snap together reusable components that scale from prototypes to production. It's how we make computer vision feel more like software engineering – and less like science experiments."

Plainsight CEO Kit Merker also provided I Programmer with an explanation of how OpenFilter can help developers build computer vision apps: 

"Open source projects help developers learn new technologies and become productive quickly. Every software problem can be solved by adding a new layer of indirection or a new abstraction, and that's precisely what we're doing with OpenFilter. Computer vision apps are really powerful, and we haven't re-invented computer vision, but rather we are helping developers to build computer vision apps in a way that is more reusable and scalable. We hope this will help more developers have access to ML-powered computer vision as a tool in their toolbox as they build new AI applications."

OpenFilter is available now under the Apache 2.0 license. There's also an Early Access Program for the commercial version of Plainsight, which includes a hardware starter kit and commercial support. 

More Information

OpenFilter On GitHub

OpenFilter Website

Embedded Vision Summit

Related Articles

What Fools AI, Fools A Human

Can An LLM Hear And See?

To be informed about new articles on I Programmer, sign up for our weekly newsletter, subscribe to the RSS feed and follow us on Twitter, Facebook or Linkedin.

Comments

Make a Comment or View Existing Comments Using Disqus

or email your comment to: comments@i-programmer.info