We at DirectAI are excited to unveil computer vision enabled live video through Ant Media.
Real-time object tracking promises transformative capabilities for businesses, from augmented robotics perception stacks to quality checks in production lines. Yet, the complexity of implementing such technology, especially for teams that lack machine learning as a core focus, is often a significant barrier.
Moreover, existing solutions struggle to balance performance with iteration speed, presenting significant challenges for real-time applications. They often require significant upfront investment through data assembly & training or only work on a specific set of objects.
To address this, DirectAI, in collaboration with Ant Media, is revolutionizing how businesses extract information from live video streams, democratizing access to powerful, easy-to-implement computer vision technology.
Ant Media’s custom server is the foundation for video applications in 130+ countries around the world. Complementing this, DirectAI enables rapid deployment of models defined in plain language, turning them into actionable insights within seconds.
DirectAI’s tech is specifically designed to meet the stringent latency and throughput demands of live video applications.
Together, Ant Media and DirectAI provide a unique and powerful solution, transforming the way businesses utilize information in live video streams, regardless of their machine learning expertise or resource availability.
Building Next-Generation Computer Vision
DirectAI has built a system that skips the model training process entirely through large Vision-Language Models trained on billions of image-caption pairs. All you need to build a model is a plain-language description (e.g. “hat”).
We work with frames – either examining them in sequence to track objects across time, detecting sets of objects in an individual image, or categorizing images into classes.
And DirectAI isn’t just using off-the-shelf machine learning. Our proprietary methods enable users to not only define objects in text, but also iterate quickly through edge-case definitions. If the “hat” model fails to detect “fedora” in testing, simply add “fedora” to the “positive” list. If the “hat” model mistakes an “afro” for a hat, simply add “afro” to the “negative” list. And if our model doesn’t understand your text description, you can just show it a single example!
DirectAI & Controllability
Ant Media has made the partnership process incredibly smooth. The Ant Media marketplace enables applications that have been built on top of the Ant Media Server to be monetized.
One of DirectAI’s early customers needed custom software built on top of their Ant Media server, so releasing that same work as a plugin to the public was a natural choice. What’s more, Ant Media has open-sourced a number of simple plugin templates that enabled DirectAI to extract and export stream frames for inference.
Check out a demo of DirectAI’s plugin below:
Why Computer Vision Matters for Your Application
DirectAI’s technology offers a significant breakthrough for teams that would otherwise be unable to implement computer vision due to engineering capacity or data constraints. There are a number of real world problems we’re excited to help tackle, namely:
- Identifying obstacles or objectives for robots/drones
- Identifying harmful material (e.g. weapons or nudity) in a trust & safety context
- Detecting defects on manufacturing lines
- Anonymous counting people
- Indexing video for search
We’d love to talk to you about integrating computer vision into your Ant Media stack. Please schedule time on our calendar or reach out to ben@directai.io so we can fast track your onboarding. Our Ant Media Marketplace listing is linked here.
If you want to learn more about DirectAI’s internal technology, feel free to check out our publicly accessible classification/detection API and quickstart repo.
