The full AI lifecycle @ Embedded World
When to use iterative development? You should use iterative development only on projects that you want to succeed.
Machine Learning and Artificial Intelligence will propel industrial vision applications into the future. AI enables hitherto unseen accuracy in product inspection, allowing for accurate detection of a variety of production deficits. AI allows for accurately recognizing and counting changing SKUs as they are presented on ever changing assembly lines. AI allows for segmenting images beyond borders to provide an unparalleled insight into the production process. During Embedded World we will, together with Edge Impulse and Advantech, demonstrate the effortless no-code creation and iteration of embedded edge AI applications for industry.
Edge AI in industry
Artificial Intelligence in its essence allows computers to learn advanced tasks based on examples; provide a computer with sufficient examples of a product and, over time, the computer can learn to tell the good from the bad. This used to be an engineering challenge only available in highly constrained and highly costly proofs-of-concept. However, in recent years AI hardware, AI model training, and AI deployment have advanced to a level where robust AI solutions can be created effortlessly and rapidly. And, AI can be deployed securely to edge devices: by moving an AI model close to the data, latency is reduced, network costs are mitigated, and data is as secure as it can be.
- Classify (varying) SKUs passing by on a conveyor belt; Use classification for accurate product counting and measurement of change-over times.
- Segment complex images to check product quality; we have supported cases ranging from bark detection on trees for the paper industry to detecting anomalies in the caps of vaccine bottles.
- Asset and Parcel tracking; Using advanced QR and bar-code recognition combined with AI to inspect package quality.
Creating the applications above is easier than ever before. However, most applications demand iteration: off-the-shelf applications often need to be (re-)trained to satisfy the use-case at hand. The power of the combination of Edge Impulse (effortless AI model training and labeling), Scailable (full AI pipeline deployment and management to edge devices), and Advantech (sturdy industrial hardware optimized for AI in industry) is to enable an effortless AI lifecycle to create highly robust and accurate AI solutions for your use-case.
The Edge AI lifecycle
Creating AI used to be hard, almost impossible. Hundreds of thousands of examples were necessary to train a “supercomputer” to come close to the desired accuracy. However, over the last decades technology has developed rapidly. All throughout the Edge AI lifecycle the process has now become feasible:
- Annotation of training data: The Edge Impulse platform makes the handling of training data as easy as it can be. Simple visual annotation of training images that is supported by highly advanced active learning methods to make labelling accurate and fast.
- Model training and testing: The Edge Impulse platform allows for extremely simple model training and evaluation. Using the Edge Impulse FOMO model architecture allows you to create (through transfer learning) highly accurate segmentation models while only requiring a small number of example images. And, you can immediately address how well your model works on your test data.
- Deployment target selection: Running your AI model—or more accurately, the AI pipeline from input image to output signal—requires robust and specialized hardware. Advantech offers AI ready hardware such as the Advantech ICAM-500; a single integrated industrial camera with sufficient power to run advanced segmentation models.
- Deployment and management: Preparing your model pipeline and transferring your trained model to your target device used to be a challenging embedded engineering task. Scailable provides highly optimized, remotely configurable pipelines with modular AI deployment capability. Simply purchase the ICAM with the Scailable AI manager pre-installed and there is no need for any on-device engineering.
- Iterate and improve: The faster you can iterate, the faster you can learn, and the more value you will bring. The Scailable AI manager allows you to collect new training images whenever models are uncertain, and, through the integration of your Edge Impulse account with the Scailable AI manager you can easily retrain your model and re-deploy.
We will provide a live demonstration of the ease by which one can train and re-train AI models, deploy complete pipelines without writing code, and create robust AI solutions in industry.
The right tools for the job
As with every task, creating high quality solutions requires the right tools for the job. Edge AI solutions need appropriate hardware targets, model creation, and model deployment and management to become truly valuable.
During Embedded World we will demonstrate the use of the novel Advantech ICAM-500 for industrial applications. The Advantech ICAM-500 series is a highly integrated Industrial AI Camera that reduces installation and maintenance effort significantly. The device is equipped with programmable, variable focus lenses, LED illumination, a SONY industrial grade image sensor, multiple core ARM processors, and an NVIDIA AI system on the module.
The presented setup at Embedded World is also available for alternative Advantech devices such as the Advantech ICR-4xxx industrial router series and the Advantech MIC-AI series. The Advantech ICAM can be purchased directly with the Scailable AI manager installed.
Edge Impulse: training and annotation
Models are core to any successful edge AI application. Model training used to be cumbersome but Edge Impulse provides data capturing, data annotation and advanced model training (with exceptional performance both in model accuracy and runtime) within minutes. The intuitive platform takes you from start to end in AI model creation. The Edge Impulse platform is easily integrated with the Scailable AI deployment platform to move trained models directly to your edge device.
Scailable: large scale, effortless deployment
AI models and hardware need to come together to facilitate the full AI pipeline. The Scailable AI manager, installed on selected Advantech hardware, provides a highly efficient AI pipeline from start-to-end that can easily be remotely configured. The Scailable platform ensures robust and secure deployment at scale, saves valuable on device engineering, and—integrated with Edge Impulse—allows for extremely fast iterations.
We are looking forward to meeting you during Embedded World. You will be able to see our demonstration in Hall 3, Booth 3-339, where we are joining Advantech. From our side, Dominique van Doorn and Wim de Wispelaere will be present at the booth; feel free to reach out and schedule an appointment (or walk by any time).
- Learn more about our embedded world demo by visiting our Embedded World page.
- Try out edge AI for yourself by installing the Scailable AI manager, and connecting your Scailable account to your Edge Impulse account.
- Or, if you want t0 get started at scale, contact our sales to kickstart your edge AI solution.
Why We Are Joining Network Optix
Today, the entire Scailable team is joining Network Optix, Inc., a leading enterprise video platform solutions provider headquartered in Walnut Creek CA, with global offices in Burbank CA, Portland OR (both USA), Taipei Taiwan (APAC HQ), Belgrade Serbia, and shortly in Amsterdam (EU HQ). Network Optix, since its founding, has set out to “solve” video […]
Scailable supporting Seeed NVIDIA devices
We are excited to announce support for Seeed’s NVIDIA Jetson devices. The AI manager, and all our edge AI development tools, can now readily be used on Seeed devices. As we all know, edge AI solutions often start from creative ideas to optimize business processes with AI. These ideas evolve into Proof-of-Concept (PoC) phases, where […]
From your CPU based local PoC to large scale accelerated deployment on NVIDIA Jetson Orin.
Edge AI solutions often start with a spark of imagination: how can we use AI to improve our business processes? Next, the envisioned solution moves into the Proof-of-Concept (PoC) stage: a first rudimentary AI pipeline is created, which includes capturing sensor data, generating inferences, post-processing, and visualization of the results. Getting a PoC running is […]