New native video annotation capabilities include, entity persistence, 4K resolution support, and 50% more efficient labeling
Austin, TX May 12, 2020:
Alegion, the industry-leading data labeling platform for machine learning, today announced the release of its next-generation video annotation solution, which includes support for long-running sequences, native 4K video support, object tracking, and ML-driven object proposal.
Alegion’s video annotation solution is targeted at data science teams that are building object tracking algorithms that identify and track individual objects of interest over time. By supporting an unlimited number of types and attributes of objects, even the most complex tracking challenges can be addressed. Also called “entity persistence”, this capability is needed for algorithms that track specific instances over time even if an entity leaves and reenters the frame.
Unlike other offerings, Alegion’s annotation platform natively supports video as a distinct data type. Many other solutions treat videos as individual images that are annotated and then recombined, making object tracking over time untenable. The platform manages loading subsets of a video at a time, allowing the video to remain intact for labeling so annotators retain vital context for recognizing people and objects as they go in and out of frame in the same clip. Additionally, Alegion’s solution supports annotation of videos of up to 4K resolution for companies requiring the fidelity provided by high-resolution sensors.
The new video annotation capabilities expand the embedded machine learning capabilities within the Alegion platform to include ML-driven pre-labeling. A video is first processed by a computer vision model and labels identifying known targets are proposed and localized on each frame. Human annotators then review and correct as needed, improving the machine-based proposals over time and allowing manual annotation to focus on the most difficult cases. Pre-labeling has shown to reduce the annotation effort by as much as 50% and can be used with other features like automated instance selection and interpolation to further reduce manual annotation.
“Video annotation has become increasingly important in training computer vision models due to the temporal context it provides,” said Chip Ray, CTO of Alegion. “Scene understanding, real-time interaction systems, and other perception model advances are only possible through video analysis. The emergence of 4K sensors, advances in drone technology, and sensor fusion has brought in new use cases from multiple industries including retail, tech, agriculture, and autonomous vehicles. Our customers are applying this technology to retail self-checkout, automated surveillance, and visual inspection. While there are various ways to annotate video, there aren’t solutions to maintain context and track targets through long-running sequences. This ability and the efficiencies gained through ML automation give Alegion the most advanced and comprehensive video annotation solution in the market,” Chip continued.
The demand for labeling continues to grow with the increasing investment in machine learning as tracked by leading analyst firm Cognilytica. Kathleen Walch, Managing Partner and Principal Analyst for Cognilytica, stated, “Video annotation presents unique challenges due to complex annotation requirements and the large amounts of data that must be processed. The most effective data labeling solutions approach video as a distinct data type and build specific technology to make annotation more efficient and accurate. Alegion’s solution appears well suited to address the rapidly evolving annotation space.”
Alegion’s video annotation is now available for all customers as part of the Alegion platform. To learn more and see a demo, visit: https://www.alegion.com/video-annotation. Alegion will be showing platform capabilities as a part of the AI Demo Showcase on May 21, 2020, which will be available for re-play.