Our CTO Tamar Shoham spoke with @AutoSens_ TV about ML-Safe compression: -> Why we can't ignore the elephant sitting on the servers, and how to identify the information that matters least to the model. -> Why lossless and content-adaptive compression must be safe for machine vision models. -> How we benchmarked ML-Safe compression across 2D and 3D object detection, depth estimation and reasoning models. piped.video/watch?v=IBMasQdx…

On stage at the @dSPACEglobal User Conference, talking about ML-Safe Lossless compression. 12-bit Bayer in, 12-bit Bayer out, zero data loss, in real time. Running on the GPU's video encoder that is already in the data logger (47% smaller than raw in our initial tests). Test it on your video data: beamr.com/lossless

Lossless compression for RTMaps: 12-bit Bayer in, 12-bit Bayer out, in real time. It runs on the GPU already in the data logger, while the compute cores keep running without extra load. Capture more in the vehicle, and move it faster and at lower cost. RTMaps users will get two modes to choose from: -> Lossless where you need the outputs bit-exact. Initial tests showed 47%. -> Content-adaptive, for up to 50% smaller files, confirmed against the model's own accuracy metrics. Beamr and @INTEMPORA will demonstrate it this week: -> @AutoSens_ Europe in Barcelona, September 22–24, stand 131 -> @dSPACEglobal User Conference West Coast, September 23 in Santa Clara, California Visit beamr.com/lossless

Your test vehicle comes back from a shift and the recorder is full. It generates more video than you can store or move. A single vehicle often runs 11 cameras, sometimes 16, and can record tens of terabytes in one shift. Much of it is recorded 12-bit Bayer, and the encoder in the vehicle can't compress that format, so the footage stays raw. Now someone has to decide: keep the raw footage and pay for it in drives, swaps and upload time, or delete video data and hope a future model never needs it. We heard that from AV and ADAS teams we worked with - so we built the solution. Today we launched Beamr ML-Safe lossless compression. It takes 12-bit Bayer raw video data and compresses it bit-exact: -> Running on the video encoder that is integrated in the GPU already in the data logger. -> The encoder runs separately from the GPU compute cores - so it draws on capacity the AI models are not using. -> In our testing on 12-bit Bayer recordings from 8-megapixel cameras, compressed files were 47% smaller than raw. Test it on your video data, or meet us at Barcelona, AutoSens Europe, September 22–24, stand 131: beamr.com/lossless

We won! Best of show award @IBCShow by TV Tech. AI Quality enhancement, verified on your content by real viewers in days: Low-resolution source -> NVIDIA Video Super Resolution with Beamr content-adaptive bitrate, up to 50% lower bitrate -> High-resolution, live channel or media archive -> Validate with Beamr VISTA, by real viewers on your content. Talk to us about your upscaling plan or whatever's stuck in your pipeline beamr.com/ibc26

Take the Metro to IBC -> See our Ad -> Take a Selfie -> Come to our Booth -> Get the Shirt! Your viewers expect 4K - so give them 4K beamr.com/ibc26 Stand 1.D22, Hall 1 @IBCShow

Demos of super resolution can be easy. Getting it into broadcast production is the hard part. We've been running proofs of concept with leading media companies on super resolution - HD upscaled to 1080p or 4K, live or media library, inside real delivery chains. The same questions came up every time: ? Which content actually improves with AI enhancement, and which doesn't ? Where it fits in the delivery chain ? Whether the economics hold once 4K detail has to fit your bitrate So we designed Beamr Blueprint to settle them before production with a decision-ready plan. Mapping your pipeline, source to screen, we pinpoint where enhancement fits, and where it doesn't we say so. All is tested and verified on your footage and with real viewers - who are the only ones to actually see if the AI-enhanced content looks better. Talk to us about Beamr Blueprint or whatever's stuck in your pipeline. See our demos of AI enhancement with @NVIDIAAI and @VAST_Data. @IBCShow Stand 1.D22, Hall 1, September 11–14: beamr.com/ibc26

If you work with AV/ML data, you’ve probably asked yourself this at some point: How much can I compress before the model notices? That’s *exactly* what we set out to test. In our upcoming webinar, we’ll share real-world case studies across multiple computer vision models and tasks, and what they tell us about building more efficient AV/ML data pipelines. Take the first steps toward ML-Safe Compression 🚗 Register now >> zoom.us/webinar/register/701…

At @IBCShow we will present a demonstration with @VAST_Data enabling broadcasters and content owners to enhance, reuse and monetize their media libraries, while keeping the resulting storage and delivery economics under control. Common quality metrics can only tell you how close the output is to the source. Only viewers can tell you whether it actually looks better. Beamr VISTA runs subjective quality tests on your own content, with real viewers, within days rather than the weeks traditional subjective testing takes. Visit our booth at IBC, Stand 1.D22 (Hall 1) - see it yourself, and tell us what you see: beamr.com/ibc26 @NVIDIAAI

This week we will demonstrate at @IBCShow a production-ready pipeline from HD up to 4K. For both live broadcasting and media archives - within existing infrastructure and without the delivery-cost penalty upscaling normally carries. Source 720p/1080p -> AI quality enhancement or upscaling with NVIDIA Video Super Resolution and Beamr's content-adaptive compression -> Standard output AVC/HEVC/AV1 And then proof that the AI actually made the footage look better. Beamr VISTA runs subjective quality tests on your own content, with real viewers, within days. Read the preview: shorturl.at/B0uqR Meet Beamr video experts at IBC, booth 1.D22: beamr.com/ibc26

4K live is often not produced in native 4K. It is upscaled automatically by the television set - the last device in the chain, with no production quality control. Beamr pipeline powered by NVIDIA moves that decision back into the broadcaster’s hands: -> NVIDIA Video Super Resolution upscales the HD feed -> Beamr content-adaptive encoding delivers streams up to 50% smaller -> The pipeline runs end-to-end on NVIDIA RTX PRO GPUs -> Beamr VISTA validates output quality on your content, with real viewers, in days “AI is transforming how live content is produced and delivered,” said Richard Kerris, GM and VP of Media & Entertainment at NVIDIA. “With NVIDIA Video Super Resolution, accelerated computing and Beamr’s content-adaptive encoding, broadcasters and streaming providers can enhance existing HD video and efficiently deliver higher-quality viewing experiences across cloud and on-premises workflows.” -> See the live demonstration at IBC 2026, Stand 1.D22 (Hall 1). Book a meeting: beamr.com/ibc26 @NVIDIAAI

"Is it safe to compress this footage, or will compression change what the models see, and degrade the footage for the tasks that rely on it?" Every AV team hits that question. Too often the answer is to keep everything raw and figure it out later. The instinct not to touch the data is understandable, but at petabyte scale it stops working. You either pay for storage you can't handle, or you end up deleting footage you'll need later. In a new blog post, we discuss what makes compression ML-Safe: when the change it causes in model behavior is statistically indistinguishable from the variance the model already tolerates due to camera noise or other real-world permutations. -> Read the blog: shorturl.at/VgsSV

Is your video data pipeline becoming the bottleneck? Vehicle storage, I/O time, training throughput, cost — all of it, at once. Let's talk about your AV pipeline: Evaluate it end to end with our video data experts. Delete data you may need, or risk your models — those were the options. Now there's a plan your team can act on with Beamr Blueprint. Visit beamr.com/blueprint_av

Managing petabyte-scale video data is hard, and autonomous vehicle teams struggle to handle it from capture to training. That's why the RTMaps AI Store is the right place for our ML-safe compression technology. Where the teams already are, on the pipelines they already use, with the clear advantage of reducing video data size while preserving the ML accuracy the models depend on. Try it on your video data in the RTMaps AI Store intempora.com/safe-video-dat…

Autonomous vehicle datasets double annually, according to industry estimates. The massive real-world and synthetic footage slows down development and drives up cost. Just a few hours of test driving can fill a large in-vehicle video recorder. When AV teams asked us to bring our expertise directly into their pipelines, with the judgment to know where compression is safe and where it is not - we got it immediately. Beamr Blueprint AV launched today. In Blueprint, we evaluate AV/ADAS pipeline end to end. We test compression, lossy or lossless, on specific models, data, and KPIs - delivering a verified plan of ML-safe video data built for your stack. The first Blueprint engagement is already underway with a global AV program. Ready to test your AV video data pipeline? Visit beamr.com/blueprint_av

𝐋𝐞𝐬𝐬 𝐯𝐢𝐝𝐞𝐨 𝐝𝐚𝐭𝐚. 𝐒𝐚𝐦𝐞 𝐌𝐋 𝐚𝐜𝐜𝐮𝐫𝐚𝐜𝐲. 𝐍𝐨𝐰 𝐢𝐧 𝐑𝐓𝐌𝐚𝐩𝐬 ✨ We're excited to announce that Beamr's ML-safe video data technology is now available on the RTMaps AI Store - making it the first compression technology on the platform. Autonomous vehicle programs generate tens to hundreds of petabytes of video data, all of which must be stored, transferred, and processed by AI perception models. Through our partnership with @Intempora, a dSPACE company, Beamr's content-adaptive bitrate (CABR) technology integrates directly into RTMaps, reducing video data by up to 50% while preserving the information AI models rely on. Same pipeline. Same model behavior. Up to 50% less video data. Read the full announcement → globenewswire.com/news-relea…

@BeamrVideo is raising the bar for live sports streaming with NVIDIA RTX Video Super Resolution, enabling premium 4K/1080p experiences that are quality-validated, bitrate-efficient, and ready to deliver without rethinking the entire workflow.