Beamr Imaging has launched a new lossless compression feature for its ML-Safe solution, designed to reduce 12-bit Bayer video file sizes by 47% for autonomous vehicle test fleets. The technology utilizes existing GPU video encoders to preserve raw data integrity without requiring additional hardware.
The immediate implication is a significant reduction in the physical and financial overhead of managing tens of terabytes generated per shift by autonomous test vehicles. By doubling data-capture capacity on existing hardware, Beamr allows engineering teams to retain high-fidelity raw footage that is critical for training future machine learning models. Within the broader ecosystem, this technology bridges the gap between massive sensor data generation and the practical limits of cloud upload speeds and drive swap logistics. Watch for adoption rates among ADAS companies following the AutoSens Europe demonstration to see if this becomes a standard component in the autonomous vehicle data pipeline.
Beamr Imaging has built its ML-Safe product line around a specific niche: preserving raw sensor data integrity for machine learning training pipelines while reducing storage and transfer costs. The company's broader strategy targets autonomous vehicle developers and ADAS teams that generate terabytes of multi-camera footage per shift. Beamr's existing customer base spans broadcast and post-production, where its Beamr 5 codec has been deployed for high-efficiency encoding, but the ML-Safe line represents a deliberate pivot toward the automotive and robotics data pipeline market. At AutoSens Europe 2026, Beamr demonstrated the lossless compression capability alongside its existing ML-Safe lossy offering, positioning the combined solution as a way to maximize data-capture density on existing GPU hardware without additional compute investment.
The business case for lossless compression in autonomous vehicle fleets centers on the economics of data retention and cloud ingestion. Autonomous test vehicles typically generate between 5 and 20 TB of raw sensor data per day, and companies must balance the cost of storing everything against the risk of discarding footage that could improve model performance. Beamr's approach of running lossless compression on existing GPU encoders in data loggers avoids the capital expenditure of dedicated compression hardware, which is a meaningful differentiator for fleet operators managing hundreds of vehicles. The company has not disclosed specific OEM or Tier-1 partnerships tied to this launch, but the product targets the same buyers evaluating solutions from competitors in the automotive data pipeline space.
In the broader encoding and compression landscape for autonomous vehicle data, Beamr competes with both general-purpose and specialized approaches. NVIDIA's Drive platform includes hardware-accelerated compression capabilities integrated into its Orin and Thor SoCs, while companies like Hailo and Mobileye handle sensor data compression as part of their perception stacks. For the specific use case of lossless Bayer data reduction, Beamr's software-only approach that leverages existing GPU video encoders offers a path to deployment without hardware refresh cycles. The 47% reduction figure places Beamr's lossless mode in a range that could meaningfully reduce cloud upload times and storage costs at fleet scale, particularly for companies running continuous data collection programs across large vehicle fleets.
Beamr has launched ML-Safe, a lossless compression solution that reduces 12-bit Bayer video file sizes for autonomous vehicle fleets by 47%. By utilizing existing GPU encoders, the software allows engineering teams to double data-capture capacity without hardware upgrades, ensuring high-fidelity raw footage is preserved for critical machine learning model training.
Beamr ML-Safe is a software-based lossless compression solution designed to reduce the file size of 12-bit Bayer raw video output from autonomous vehicle camera sensors.
The technology reduces 12-bit Bayer video file sizes by 47% while remaining bit-exact, meaning no raw data integrity is lost.
No, the solution utilizes existing GPU video encoders already present in data loggers, meaning no additional hardware is required to process the video streams.
It eliminates the trade-off between high storage costs and deleting valuable training data, allowing teams to retain more high-fidelity footage for training machine learning models.
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source
