Training an AI model on video data processed by Beamr’s content adaptive technology reduced depth estimation error by 30.7% for safety-critical road users, including pedestrians and motorcyclists, and improved the model’s resistance to compression.
HERZLIYA, Israel, May 6, 2026 (Globe Newswire) — Beamr Imaging Ltd. (NASDAQ: BMR), a leader in video optimization technology and solutions published research demonstrating that machine vision models fine-tuned on video compressed by Beamr’s patented Content Adaptive Bitrate (CABR) technology are more resilient than models trained on uncompressed data, while simultaneously reducing the amount of video data on which machine vision development relies.
Machine vision teams working with petabytes of video data for autonomous vehicles (AVs) and other video AI applications typically consider compression as a process to manage this scale. Our findings reframe adaptive compression as an asset that enhances the resiliency of AI models, with the benefit of reducing storage and network costs and infrastructure. This study extends Beamr’s ML-Safe benchmark to validate the potential performance assets of AI models trained across machine vision applications.
In this study, we evaluated Depth Anything V2, a state-of-the-art monocular depth estimation model. The model was fine-tuned based on AV video data compressed with Beamr’s technology, resulting in a 35.2% reduction in file size compared to baseline compression. The fine-tuned model demonstrated a 30.7% reduction in depth estimation error for vulnerable road users, including pedestrians and motorcyclists, and a total reduction of 16.0% across all object classes. The full methodology and results are available in our blog post.
“This study shows that compressed video data can produce more robust models, rather than less robust ones.”
Beamr CPO Dani Megrelishvili said:
“This shows that compression plays a different role in our customers’ pipelines, from the cost they tolerate to the tools they deploy.”
“Machine vision teams face a structural trade-off: Compress video data to manage scale, or face the increasing cost and infrastructure challenges of running AI models without compression.”
Ronen Nissim, ML Lead at Beamr, said:
“Our research shows that this trade-off is more flexible than the industry assumed. By using compressed footage as an extension during fine-tuning, we created a model that performed better on the validation set than an equivalent model trained on uncompressed data.”
Beamr’s ML-Safe benchmark has previously validated content-adaptive compression across the AV development pipeline. Benchmarks have demonstrated up to 50% reduction in file size while maintaining object detection accuracy of 0.96 on average while maintaining high fidelity in detection, localization, and reliability consistency. Subsequent testing of the captioning workflow on the World Foundation model pipeline showed file size reductions of 41% to 57% with no measurable impact on pipeline output.
To run Beamr compression on your data, visit beamr.com/autonomous.
About Beamle
Beamr (Nasdaq: BMR) is the world leader in content-adaptive video compression, trusted by top media companies such as Netflix and Paramount. Beamr’s Perceptual Optimization Technology (CABR) is backed by 53 patents and has won an Emmy Award in the Technology and Engineering category. This innovative technology reduces video file size by up to 50% while maintaining quality and enabling AI-powered enhancements.
Beamr powers efficient video workflows across high-growth markets such as media and entertainment, user-generated content, machine learning, and autonomous vehicles. Flexible deployment options include on-premises, private cloud, or public cloud, with convenient availability for Amazon Web Services (AWS) and Oracle Cloud Infrastructure (OCI) customers.
For more information, please visit www.beamr.com or our investor website www.investors.beamr.com.
Forward-looking statements
This press release contains forward-looking statements that involve significant risks and uncertainties. Forward-looking statements in this communication may include, among other things, statements regarding Beamr’s strategic and business plans, technology, relationships, business objectives and expectations, the impact of trends and interests on our business, intellectual property or products, future results, operations, financial performance and condition. All statements, other than statements of historical fact, contained in this press release are forward-looking statements. Forward-looking statements contained in this press release include forward-looking statements such as “anticipates,” “believes,” “contemplates,” “may,” “estimates,” “expects,” “intends,” “seeks,” “may,” “might,” ” “plans,” “may,” “anticipates,” “plans,” “targets,” “aims,” “should,” “will,” or, however, not all forward-looking statements contain these words. Forward-looking statements are based on our current expectations and are subject to inherent uncertainties, risks and assumptions that are difficult to predict. Additionally, certain forward-looking statements are based on assumptions about future events that may not prove to be accurate. For a more detailed description of the risks and uncertainties affecting us, please refer to our reports filed from time to time with the Securities and Exchange Commission (the “SEC”). These include, but are not limited to, the risks detailed in the Company’s Annual Report filed with the SEC on February 26, 2026 and subsequent filings with the SEC. The forward-looking statements contained in this announcement are made as of the date hereof, and we undertake no obligation to update such information except as required by applicable law.
Investor contact information:
[email protected]
