Blogs on Plumerai Blog

Multi-Camera Re-Identification solves notification overload

Thu, 02 Apr 2026 00:00:00 +0000

Your kids are playing basketball on the driveway. You’re packing your car before a trip. Or, as in the video below, your gardener is mowing the lawn and triggers 34(!) push notifications on your phone in less than an hour.

Notification overload is the number one annoyance for smart home camera users. It’s the key reason many users have only one camera deployed at home. Because more cameras generate even more useless notifications.

Plumerai Advanced Motion Detection

Thu, 19 Mar 2026 12:00:00 +0100

Motion detection is one of the most common tasks for smart home cameras: determining whether something is moving in the scene and if so, where. It can be useful on its own, triggering a notification or starting a recording, but it also plays an important role as the first stage of a larger video intelligence pipeline. Algorithms such as people detection and face identification can operate without it, but they perform better when motion detection indicates where they should focus.

Not-quite-ivory towers

Mon, 16 Mar 2026 00:00:00 +0000

by Alex Chebykin

“How to make the damn thing work?” is the key question guiding most engineers. To answer it, reality has to be faced head-on. As you test your ideas against it, you get to watch again and again “the slaying of a beautiful hypothesis by an ugly fact”. Incidentally, this is the phrase Thomas Huxley used to describe “the great tragedy of Science”. This tragedy is clearly shared with engineering.

Meet Plumerai at ISC West 2026

Tue, 10 Mar 2026 08:33:38 +0100

Attending ISC West?

At ISC West, we’re showing how Plumerai combines tiny AI models with powerful LLMs to deliver better AI for cameras, NVRs, and VMS platforms, enabling new capabilities such as natural-language video search and other advanced AI features.

Our customers integrate the Plumerai software to ship faster, more accurate, and more efficient AI features across their product lines, while reducing cloud compute costs by 10x or more. Our neural networks are highly efficient and can run directly on the camera, on-prem, or in the cloud. When deployed in the cloud, they require far less compute than alternatives, significantly reducing cloud infrastructure costs.

Plumerai raises $8.7M Series A to connect Vision LLMs to trillions of edge devices

Tue, 16 Sep 2025 00:50:04 +0200

Press release

London – September 16, 2025 - Plumerai™, a pioneer in on-device AI for cameras, today announced an $8.7M Series A funding round, led by new investors Partech and OTB Ventures, with support from Acclimate Ventures and existing investors. This brings the total funding received to over $17M. The new funding will drive Plumerai’s ambition to enable trillions of intelligent devices, a future that is now accelerated by its new Vision LLM features.

Instantly search hours of video

Wed, 02 Apr 2025 08:20:33 +0000

Smart home cameras capture hours of footage, but who has time to sift through it all?

With Plumerai AI Video Search, you can search for anything:

“Kid on a scooter wearing a red helmet” or “Man in a blue jumper walking a labrador”

Just type it. Get results instantly.

Powered by Plumerai’s Vision LLM, it’s so efficient, it can even run entirely on the camera itself - no cloud needed, zero compute cost.

Fast & accurate Familiar Face Identification, even from a distance

Tue, 25 Mar 2025 08:20:33 +0000

Run Plumerai’s Familiar Face Identification on your video doorbells to deliver fast, accurate detection of friends and family. Enable your users to know when a loved one arrives home, or to turn off alerts for familiar faces to minimize unnecessary interruptions; it even detects them from a distance. Best of all, since all the AI runs inside the camera, privacy is respected and there’s no cloud compute cost.

The complete tiny AI solution for smarter homes

Tue, 25 Feb 2025 08:20:33 +0000

Plumerai’s complete tiny AI solution powers advanced features, including Familiar Face Identification and AI Video Search, directly on your smart home cameras and video doorbells. It’s the most accurate and compute-efficient solution on the market, all while running entirely on the edge, eliminating the need for costly cloud compute. Give your users the ability to enjoy precise and reliable notifications without compromising their privacy.

See our Smart Home Cameras page for more information and demos.

Meet Plumerai at CES 2025

Tue, 10 Dec 2024 08:33:38 +0100

CES 2025 is just around the corner! We’ll be presenting our innovations in AI video search and multi-camera re-identification, and discussing the future of smart home cameras.

Are you developing cameras? If so, send a message to hello@plumerai.com to book a meeting at our private suite in The Venetian.

Simple alerts & zero distractions

Thu, 05 Dec 2024 11:08:06 +0100

Our advanced AI software transforms your cameras into smarter, more user-friendly devices that deliver meaningful notifications tailored to your users’ needs:

Plumerai Familiar Face Identification: Empowers users to silence alerts for household members, minimizing unnecessary interruptions to their day.

Plumerai Stranger Detection: Ensures users are notified when it matters most, offering enhanced security and peace of mind.

This innovative technology filters out the noise often created by the smart home ecosystem, creating an engaging and streamlined user experience that puts the user in control. Try out Plumerai Familiar Face Identification in your browser here!

The complete on-device AI solution for smarter home cameras

Fri, 22 Nov 2024 08:20:33 +0000

Plumerai offers a complete AI solution for cameras and video doorbells, packed with advanced functionalities, from Familiar Face Identification and AI Video Search, to recognizing people, vehicles, packages, and animals!

Our tiny AI runs directly inside the camera, ensuring top-notch privacy while eliminating the need for costly cloud compute. It’s the most accurate and compute-efficient solution on the market, delivering precise notifications your users can trust. From identifying unexpected visitors to capturing animals in the garden, Plumerai ensures no important moment goes unnoticed.

Our AI models are tiny, but training them is complex and costly

Tue, 22 Oct 2024 00:09:02 +0100

This chart captures the intensity of our cloud compute usage over two weeks in September. Each colored bar serves as a reminder of the complexity and cost required to build truly intelligent camera systems.

Our AI may be tiny, but our AI factory is anything but simple. Every day, it handles thousands of tasks—training models on over 30 million images and videos, optimizing for different platforms, and rigorously fine-tuning, testing, and validating each one. All of this happens in the data center, where costs quickly skyrocket when pushing the state-of-the-art in AI.

iPod Inventor and Nest Founder Tony Fadell Backs Plumerai’s Tiny AI, Hailing It as a Massive Market Disruption Democratizing AI Technology Starting with Smart Home Cameras

Mon, 07 Oct 2024 00:50:04 +0200

Press release

London – October 7, 2024 - Plumerai™, a pioneer in on-device AI solutions since 2018, today announced a major partnership with Chamberlain Group, whose brands include myQ and LiftMaster, marking a significant milestone in the adoption of its Tiny AI technology.

To achieve features like People Detection and Familiar Face Identification, cloud-based AI and, in particular, Large Language Models (LLMs) require vast remote data centers, consume increasing amounts of energy, pose privacy risks, and incur rising costs. Plumerai’s Tiny AI can do all this on the device itself, is cost-effective, chip agnostic, capable of operating on battery-powered devices, doesn’t clog up your bandwidth with huge video uploads, and has minuscule energy requirements. Moreover, it boasts the most accurate on-device Tiny AI on the market and offers end-to-end encryption. Already running locally on millions of smart home cameras, Plumerai’s Tiny AI is making communities safer and lives more convenient, while proving that in AI, smaller can indeed be smarter. Plumerai has gained strong backing from early investor Tony Fadell, Principal at Build Collective along with Dr. Hermann Hauser KBE, Founder of Arm, and LocalGlobe.

Automatic Enrollment for Plumerai Familiar Face Identification!

Thu, 03 Oct 2024 12:09:02 +0100

We have launched Automatic Enrollment for our Plumerai Familiar Face Identification! Automatic Enrollment makes it effortless to register new individuals for identification; as soon as a person walks up to the camera, they will be automatically registered, and users can tag them with a name in the app.

This feature:

identifies people from an 8m/26ft distance (often beyond),
excels with a variety of camera positions, camera angles, and in low light conditions,
and most importantly, it has been built to run on the device (not in the cloud!), so it doesn’t compromise on privacy.

Plumerai Familiar Face Identification is now operational on millions of cameras and we are proud to share that it is the most accurate solution for the smart home (outperforming Google Nest!). Offering this new feature really takes it to the next level.

High accuracy, even in the dark

Tue, 24 Sep 2024 12:09:02 +0100

Plumerai Familiar Face Identification enables smart home cameras to notify you when loved ones arrive safely, activate floodlights if a stranger loiters in your yard, and quickly search through recordings. It’s also a critical building block for the vision LLM features that we’re developing.

As shown in the video, it performs exceptionally well even in complete darkness, accurately and effortlessly identifying people. It works perfectly from many camera positions and with wide-angle lenses.

Plumerai AI Factory

Tue, 27 Aug 2024 11:08:06 +0100

Developing accurate and efficient AI solutions involves a complex process where numerous tasks must work seamlessly together. We refer to this as our Plumerai AI factory.

Our AI factory is built on a foundation of carefully integrated components, each playing a crucial role in delivering accurate and robust AI solutions.

From data collection and labeling to model architecture and deployment, every step is meticulously designed to ensure optimal performance and accuracy. We continuously refine our algorithms and models, apply advanced training strategies, and conduct rigorous data unit tests to identify and mitigate potential failure cases.

Multi-camera smart homes are noisy!

Fri, 26 Jul 2024 08:20:33 +0000

Multi-camera smart homes enhance security, but more cameras mean more notifications. Imagine Emma arriving home to the house pictured below and walking through her home to the garden. Her household would receive three notifications in quick succession from the video doorbell, indoor camera, and outdoor camera. That’s not a good user experience!

At Plumerai, we solve this with our multi-camera Re-Identification AI, which recognizes individuals by their clothes and physique. It tracks the same person as they move around the home, reducing unnecessary alerts. When combined with our Familiar Face Identification, the notification can become even more specific. So, when Emma arrives home and moves through the house to the garden, users receive just one useful notification:

Plumerai Re-ID groups video events

Thu, 18 Jul 2024 11:08:06 +0100

With Plumerai’s Re-identification AI embedded in your smart home cameras, they can remember and recognize a person’s clothing and physique. So, you can seamlessly merge video clips of the same individual into a single, comprehensive video. The result is a simplified, uncluttered activity feed, allowing users to quickly locate and view the video clips they’re interested in, without having to open multiple short clips.

Integrated with Plumerai’s Familiar Face Identification & Stranger Detection, the system also identifies known and unknown individuals. In the example below, users will receive just one, concise video of ‘Emma’ playing outside, instead of dozens of short videos.

Plumerai Package Detection AI

Tue, 09 Jul 2024 12:09:02 +0100

Integrate Plumerai Package Detection AI into your video doorbells and give your customers the knowledge of when and where their packages have been delivered. Our AI detects a wide variety of packages, their location, and drop off or pick up. Plumerai Package Detection is delivered as part of our full smart home camera AI solution.

Simply brilliant alerts with Plumerai Re-ID

Wed, 26 Jun 2024 11:08:06 +0100

Plumerai’s breakthrough Person Re-identification AI remembers and recognizes a person’s clothing and physique, enabling it to understand when the same person is going in and out of view. Combined with Plumerai’s Familiar Face & Stranger Identification, it can also recognize known and unknown individuals. So if Emma is gardening or taking out the garbage, household members receive a single, insightful alert rather than a flood of unhelpful notifications.

Fisheye lens? No problem!

Thu, 20 Jun 2024 12:09:02 +0100

Fisheye lenses capture a much wider angle, so users can see a clear view of their porch. However, the fisheye distortion often makes accurate familiar face identification difficult. Plumerai’s AI has been designed to work with a wide variety of fields-of-view, enabling high detection accuracy even on wide angle lenses!

Plumerai Package Detection AI

Tue, 11 Jun 2024 15:20:33 +0000

Integrate Plumerai’s new Package Detection AI for video doorbells, so your customers know when packages are delivered to their home. Send snapshots of each package delivery or pickup to give them extra peace of mind.

Plumerai Package Detection AI integrates seamlessly with Plumerai People, Animal, Vehicle, and Stranger Detection, as well as Plumerai Familiar Face Identification.

Accurate Stranger Detection AI

Thu, 06 Jun 2024 15:20:33 +0000

Plumerai is enhancing stranger detection! 🚀

Unlike other systems that flag ‘Stranger’ when no face match is found, our AI goes a step further. Plumerai’s technology can distinguish between unclear or distant faces and actual strangers. When the face is unclear, our algorithm labels it as ‘Identity Unknown’ rather than jumping to conclusions. As a result, you can rely our technology to power accurate notifications and security responses, such as turning on floodlights at night and activating alarms when you’re not at home.

Intelligent floodlights with Plumerai Stranger Detection AI

Wed, 17 Apr 2024 15:20:33 +0000

We’re taking smart home cameras to the next level with Plumerai’s Stranger Detection AI! Our technology accurately distinguishes between familiar faces and strangers, allowing your customers to tailor their alerts and security response. For an advanced smart home experience, link the doorbell to the home security system: welcome loved ones with warm lighting, and deter unwelcome visitors with red floodlights and an alarm.

No more "Did you get home ok?"

Wed, 20 Mar 2024 15:20:33 +0000

Offer peace of mind to your customers by deploying Plumerai’s Familiar Face Identification AI to your video doorbell. Our technology recognizes familiar faces, offering personalized alerts like “Jane arrived home”. The best part? Plumerai AI models run on the edge, so your customers’ images don’t leave the device, ensuring their privacy.

Try out our browser-based webcam demo here: plumerai.com/automatic-face-identification-live

Plumerai’s highly accurate People Detection and Familiar Face Identification AI are now available on the Renesas RA8D1 MCU

Fri, 15 Dec 2023 09:12:58 +0100

Discover the power of our highly efficient, on-device AI software, now available on the new Renesas RA8D1 MCU through our partnership with Renesas. In the video below, Plumerai’s Head of Product Marketing, Marco, showcases practical applications in the Smart Home and for IOT, and demonstrates our People Detection AI and Familiar Face Identification AI models.

The Plumerai People Detection AI fits on almost any camera with its tiny footprint of 1.5MB. It outperforms much larger models on accuracy, and achieves 13.6 frames per second on the RA8’s Arm Cortex-M85. Familiar Face Identification is a more complex task and requires multiple neural networks running in parallel, but still runs at 4 frames per second on the Arm Cortex-M85, providing rapid identifications.

Enable secure end-to-end encryption with Plumerai AI at the edge

Fri, 24 Nov 2023 11:08:06 +0100

To offer AI-powered alerts and recordings, smart home cameras typically send unencrypted footage to the cloud for AI analysis. This is because if the video is encrypted, the AI can’t analyze it; however, sending unencrypted data to the cloud poses risks for privacy and cyber attacks. With Plumerai, all the AI occurs on the camera, enabling you to offer end-to-end encryption on your smart home cameras, as well as advanced AI features! Plumerai gives you the best of both worlds without compromise.

No more notification overload

Thu, 09 Nov 2023 15:20:33 +0000

Say goodbye to irrelevant notifications from your video doorbell. No more notifications from moving branches, shadows, or of yourself coming home. Plumerai’s technology enables your video doorbells to accurately detect people & vehicles and even recognizes household members. So you are only alerted about the events that matter to you.

Try it in your browser now: plumerai.com/automatic-face-identification-live

Welcoming familiar faces. Keeping watch for strangers.

Thu, 26 Oct 2023 15:20:33 +0000

Upgrade your smart home cameras with Plumerai’s Familiar Face Identification AI for fewer, more relevant notifications. Our AI model runs inside the camera, so images don’t leave the device, protecting your customer’s privacy. Simply deploy our technology to your existing devices via an over-the-air software update.

Try our live AI demo in your browser: plumerai.com/automatic-face-identification-live

Arm Tech Talk: Accelerating People Detection with Arm Helium vector extensions

Mon, 02 Oct 2023 09:57:26 +0200

Watch Cedric Nugteren showcase Plumerai’s People Detection on an Arm Cortex-M85 with Helium vector extensions, running at a blazing 13 FPS with a 3.7x speed-up over Cortex-M7.

Cedric delves deep into Helium MVE, providing a comprehensive comparison to the traditional Cortex-M instruction set. He also demonstrates Helium code for 8-bit integer matrix multiplications, the core of deep learning models.

He shows a live demo on a Renesas board featuring an Arm Cortex-M85, along with a preview of the Arm Ethos-U accelerator, ramping up the frame rate even further to 83 FPS. Additionally, he showcases various other Helium-accelerated AI applications developed by Plumerai.

tinyML EMEA 2023: Familiar Face Identification

Mon, 17 Jul 2023 08:00:00 +0000

Imagine a TV that shows tailored recommendations and adjusts the volume for each viewer, or a video doorbell that notifies you when a stranger is at the door. A coffee machine that knows exactly what you want so you only have to confirm. A car that adjusts the seat as soon as you get in, because it knows who you are. All of this and more is possible with Familiar Face Identification, a technology that enables devices to recognize their users and personalize their settings accordingly.

Plumerai wins MLPerf Tiny 1.1 AI benchmark for microcontrollers again

Tue, 27 Jun 2023 12:55:04 +0000

Last year we presented our MLPerf Tiny 0.7 and MLPerf Tiny 1.0 benchmark scores, showing that our inference engine runs your AI models faster than any other tool. Today, MLPerf released new Tiny 1.1 scores and Plumerai has done it again: we still lead the performance charts. A faster inference engine means that you can run larger and more accurate AI models, go into sleep mode earlier to save power, and/or run AI models on smaller and lower cost hardware.

World’s fastest MCU: Renesas runs Plumerai People Detection on Arm Cortex-M85 with Helium

Mon, 13 Mar 2023 15:39:23 +0000

Plumerai delivers a complete software solution for people detection. Our AI models are fast, highly accurate, and tiny. They even run on microcontrollers. Plumerai People Detection and the Plumerai Inference Engine have now been optimized for the Arm Cortex-M85 also, including extensive optimizations for the Helium vectorized instructions. We worked on this together with Renesas, an industry leader in microcontrollers. Renesas will show our people detection running on an MCU that is based on the Arm Cortex-M85 processor at the Embedded World conference this week.

Plumerai People Detection AI now runs on Espressif ESP32-S3 MCU

Mon, 06 Mar 2023 11:08:06 +0100

Plumerai People Detection AI is now available on Espressif’s ESP32-S3 microcontroller! Trained with more than 30 million images, Plumerai’s AI detects each person in view, even if partially occluded, and tracks up to 20 people across frames. Running the Plumerai People Detection on Espressif’s MCU enables new smart home, smart building, smart city, and smart health applications. Tiny smart home cameras based on the ESP32-S3 can provide notifications when people are on your property or in your home. Lights can turn on when we get home and the AC can direct the cold airflow toward you. The elderly can stay independent longer with sensors that notice when they need help. Traffic lights notice automatically when you arrive. In retail, customers can be counted for footfall analysis, and displays can show more detailed content when customers get closer to them. The Plumerai People Detection software supports indoor, outdoor, low light, and difficult backgrounds such as moving objects, and can detect at more than 20m / 65ft distance. The Plumerai People Detection runs completely at the edge and all computations are performed on the ESP32-S3. This means there is no internet connection needed, and the captured images never leave the device, increasing reliability and respecting privacy. In addition, performing the people detection task at the edge eliminates costly cloud compute. We are proud to offer a solution that enables more applications and products to benefit from our accurate people detection software.

MLPerf Tiny 1.0 confirms: Plumerai’s inference engine is again the world’s fastest

Wed, 09 Nov 2022 12:55:04 +0000

Earlier this year in April we presented our MLPerf Tiny 0.7 benchmark scores, showing that our inference engine runs your AI models faster than any other tool. Today, MLPerf released the Tiny 1.0 scores and Plumerai has done it again: we still have the world’s fastest inference engine for Arm Cortex-M architectures. Faster inferencing means you can run larger and more accurate AI models, go into sleep mode earlier to save power, and run them on smaller and lower cost MCUs. Our inference engine executes the AI model as-is and does no additional quantization, no binarization, no pruning, and no model compression. There is no accuracy loss. It simply runs faster and in a smaller memory footprint.

World’s fastest inference engine now supports LSTM-based recurrent neural networks

Wed, 02 Nov 2022 09:00:00 +0100

At Plumerai we enable our customers to perform increasingly complex AI tasks on tiny embedded hardware. We recently observed that more and more of such tasks are using recurrent neural networks (RNNs), in particular RNNs using the long-short-term-memory (LSTM) cell architecture. Example uses of LSTMs are analyzing time-series data coming from sensors like IMUs or microphones, human activity recognition for fitness and health monitoring, detecting if a machine will break down, and speech recognition. This led to us optimizing and extending our support for LSTMs and today we are proud to announce that Plumerai’s deep learning inference software greatly outperforms existing solutions for LSTMs on microcontrollers for all metrics: speed, accuracy, RAM usage, and code size.

Plumerai’s people detection powers Quividi’s audience measurement platform

Fri, 30 Sep 2022 11:18:00 +0100

We’re happy to report that Quividi adopted our people detection solution for its audience measurement platform. Quividi is a world leader in the domain of measuring audiences for digital displays and retail media. They have over 600 customers analyzing billions of shoppers every month, across tens of thousands of screens.

Quividi’s platform measures consumer engagement in all types of venues, outside, inside, and in-store. Shopping malls, vending machines, bus stops, kiosks, digital merchandising displays and retail media screens measure audience impressions, enabling monetization of the screens and leveraging shopper engagement data to drive sales up. The camera-based people detection is fully anonymous and compliant with privacy laws, since no images are recorded or transmitted. With the integration of Plumerai’s people detection, Quividi expands the range of its platform capabilities, since our tiny and fast AI software runs seamlessly on any Arm Cortex-A processor, instead of on costly and power-hungry hardware. Building a tiny and accurate people detection solution takes time: we collect and curate our own data, design and train our own model architectures with over 30 million images, and then run them on off-the-shelf Arm CPUs using our world’s fastest inference engine software.

tinyML Summit 2022: ‘Tiny models with big appetites: cultivating the perfect data diet'

Fri, 13 May 2022 08:00:00 +0000

Although lots of research effort goes into developing small model architectures for computer vision, real gains cannot be made without focusing on the data pipeline. We already mentioned the importance of quality data and some pitfalls of public datasets in an earlier blog post, and have been further improving our data tooling a lot since then to make our data processes even more powerful. We have incorporated several machine learning techniques that enable us to curate our datasets at deep learning scale, for example by identifying images that contributed strongly to a specific false prediction during training.

MLPerf Tiny benchmark shows Plumerai's inference engine on top for Cortex-M

Fri, 06 May 2022 15:00:00 +0100

We recently announced Plumerai’s participation in MLPerf Tiny, the best-known public benchmark suite for evaluation of machine learning inference tools and methods. In the latest v0.7 of the MLPerf Tiny results, we participated along with 7 other companies. The published results confirm the claims that we made earlier on our blog: our inference engine is indeed the world’s fastest on Arm Cortex-M microcontrollers. This has now been validated and tested using standardized methods and reviewed by third parties. And what’s more, everything was also externally certified for correctness by evaluating model accuracy on four representative neural networks and applications from the domain: anomaly detection, image classification, keyword spotting, and visual wake words. In addition, our inference engine is also very memory efficient and works well on Cortex-M devices from all major vendors.

tinyML Talk: Demoing the world’s fastest inference engine for Arm Cortex-M

Tue, 04 Jan 2022 10:00:00 +0000

We recently announced Plumerai’s inference engine for 8-bit deep learning models on Arm Cortex-M microcontrollers. We showed that it is the world’s most efficient on MobileNetV2, beating TensorFlow Lite for Microcontrollers with CMSIS-NN kernels by 40% in terms of latency and 49% in terms of RAM usage with no loss in accuracy. However, that was just on a single network and it might have been cherry-picked. This presentation shows a live demonstration of our new service that you can use to test your own 8-bit deep learning models with Plumerai’s inference engine. In this talk Cedric explains what we did to get these speedups and memory improvements and shows benchmarks for the most important publicly available neural network models.

People detection for building automation with Texas Instruments

Mon, 03 Jan 2022 10:00:00 +0000

We’ve partnered with Texas Instruments. Together we’ll enable many more AI applications on tiny microcontrollers. Plumerai’s highly accurate people detection runs on TI’s tiny SimpleLink Wi-Fi CC3220SF MCU. Resource utilization is minimal: peak RAM usage is 170 kB, program binary size is 154 kB, and the detection runs in real-time on the low power Arm Cortex-M4 CPU at 80MHz. We’re controlling the lights here, resulting in a much better user experience. The product was demonstrated at CES 2022.

Squeezing data center AI into a tiny microcontroller

Thu, 09 Dec 2021 11:03:35 +0000

What’s the point of making deep learning tiny if the resulting system becomes unreliable and misses key detections or generates false positives? We like our systems to be tiny, but don’t want to compromise on detection accuracy. Something’s got to give, right? Not necessarily. Our person detection model fits on a tiny Arm Cortex-M7 microcontroller and is as accurate as Google’s much larger EfficientDet-D4 model running on a 250 Watt NVIDIA GPU.

The world’s fastest deep learning inference software for Arm Cortex-M

Mon, 04 Oct 2021 10:30:39 +0100

New: Latest official MLPerf results are in!

New: Try out our inference engine with your own model!

At Plumerai we enable our customers to perform increasingly complex AI tasks on tiny embedded hardware. We’re proud to announce that our inference software for Arm Cortex-M microcontrollers is the fastest and most memory-efficient in the world, for both Binarized Neural Networks and for 8-bit deep learning models. Our inference software is an essential component of our solution, since it directs resource management akin to an operating system. It has 40% lower latency and requires 49% less RAM than TensorFlow Lite for Microcontrollers with Arm’s CMSIS-NN kernels while retaining the same accuracy. It also outperforms any other deep learning inference software for Arm Cortex-M:

Great TinyML needs high-quality data

Tue, 17 Aug 2021 11:44:02 +0100

So far we have mostly written about how we enable AI applications on tiny hardware by using Binarized Neural Networks (BNNs). The use of BNNs helps us to reduce the required memory, the inference latency and energy consumption of our AI models, but there is something that we have been less vocal about that is at least as important for AI in the real world: high-quality training data.

To train tiny models, choose your data wisely

Deep learning models are famously hungry for training data, as more training data is usually the most effective way to improve accuracy. But once we started to train deep learning models that are truly tiny — with model sizes of a few hundred KB or less for computer vision tasks like person detection — we discovered that it is not so much the quantity but the quality of the training data that matters.

BNNs for TinyML: performance beyond accuracy — CVPR 2021 Workshop on Binarized Neural Networks

Thu, 01 Jul 2021 11:33:38 +0100

Tim de Bruin, a Deep Learning Scientist at Plumerai, was one of the invited speakers at last week’s CVPR 2021 Workshop on Binarized Neural Networks for Computer Vision. Tim presented some of Plumerai’s work on solving the remaining challenges with BNNs and explains why optimizing for accuracy is not enough.

tinyML Summit 2021: Person Detection under Extreme Constraints — Lessons from the Field

Tue, 20 Apr 2021 14:13:38 +0200

At this year’s tinyML Summit, we presented our new solution for person detection with bounding boxes. We have developed a person detection model that runs in real time (895 ms latency) on an STM32H7B3 board (Arm Cortex-M7), a popular off-the-shelf available microcontroller. To the best of our knowledge, this is the first time anyone runs person detection in real-time on Arm Cortex-M based microcontrollers, and we are very excited to be bringing this new capability to customers!

MLSys 2021: Design, Benchmark, and Deploy Binarized Neural Networks with Larq Compute Engine

Wed, 07 Apr 2021 00:05:36 +0200

We are very excited to present our paper Larq Compute Engine: Design, Benchmark, and Deploy State-of-the-Art Binarized Neural Networks at the MLSys 2021 conference this week!

Larq Compute Engine (LCE) is a state-of-the-art inference engine for Binarized Neural Networks (BNNs). LCE makes it possible for researchers to easily benchmark BNNs on mobile devices. Real latency benchmarks are essential for developing BNN architectures that actually run fast on device, and we hope LCE will help people to build even better BNNs.

tinyML Talks: Binarized Neural Networks on microcontrollers

Fri, 22 Jan 2021 12:09:02 +0100

For the past few months we have been working very hard on something new: Binarized Neural Networks on microcontrollers. By bringing deep learning to cheap, low-power microcontrollers we remove price and energy barriers and make it possible to embed AI into basically any device, even for relatively complex tasks such as person detection.

This week, we gave a presentation as part of the tinyML Talks webcast series where we explained what we had to build to make this work. We demonstrated how combining our custom training algorithms, inference software stack and datasets results in a highly accurate and efficient solution - in this case for person presence detection on the STM32L4R9, an ARM Cortex-M4 microcontroller from STMicroelectronics. This technology can be implemented to trigger push notifications for smart home cameras, wake up devices when a person is detected, detect occupancy of meeting rooms and for many more applications. This is a step towards our goal of making deep learning ultra low-power and a future where battery-powered peel-and-stick sensors can perform complex AI tasks everywhere.

XMOS and Plumerai partner to accelerate commercialisation of binarized neural networks

Wed, 01 Apr 2020 16:00:00 +0200

XMOS and Plumerai bring together their deep learning expertise across chip design and algorithms in a Binarized Neural Network capability, advancing the deployment of intelligence at the edge.

Bristol & London UK, 1 April 2020 – British technology companies XMOS and Plumerai have agreed a new strategic partnership that will support the development of binarized neural network (BNN) capabilities that enable AI to be embedded in a wide range of everyday devices efficiently at low-power and at low-cost.

The Larq Ecosystem

Tue, 24 Mar 2020 00:00:00 +0000

In our previous blog post, we announced Larq Compute Engine (LCE), our deployment solution for Binarized Neural Networks (BNNs). The combination of LCE, our training library Larq and the models in Larq Zoo forms the first end-to-end solution for anyone building applications using BNNs. In this post, we take a step back and look at this integrated ecosystem as a whole.

Continue reading on the Larq blog…

Announcing Larq Compute Engine v0.1

Mon, 17 Feb 2020 13:00:00 +0100

We believe BNNs are the future of efficient inference, which is why we’ve developed tools to make it easier to train and research these models. Our open-source library Larq enables developers to build and train BNNs and integrates seamlessly with TensorFlow Keras. Larq Zoo provides implementations of major BNNs from the literature together with pretrained weights for state-of-the-art models.

But the ultimate goal of BNNs is to solve real-world problems on the edge. So once you’ve built and trained a BNN with Larq, how do you get it ready for efficient inference? Today, we’re introducing Larq Compute Engine to tackle that problem.