Intel® AI: In Production | Success Stories

AAEON Technology

“Our strategic focus is bringing artificial intelligence on the edge but it has always been challenging to implement efficient algorithm balancing [for] the workload between the different hardware components. OpenVINO™ toolkit bridges this gap and gives us a unique tool that speeds up the path to production of our and our partners' solutions.”

— Fabrizio Del Maffeo, Vice President and Managing Director, AAEON Technology

ADLINK

Defect inspection for contact lenses is difficult and inefficient since defects of contact lens are usually non-rule based. Additionally, transparency of the contact lens increases the inspection difficulty,  accuracy, and overall manufacturing throughput. ADLINK AI contact lens inspection solution uses well-validated EOS-i6000-M Series vision system featured 9th Gen Intel® Core™ i7-9700E processor and 4x Intel® Movidius™ Myriad™ X VPUs executes onsite contact lens inspection via AI software partner-LEDA technology’s easy-to-use AI contact lens inspection software​. Adlink choses Intel because it they are affordable, cost-effective, and well-validated. With this Intel based solution, optimized with the Intel® Distribution of OpenVINO™ toolkit, contact lens manufacturers get a reliable all-defect-inspection solution for their products. Meanwhile, the manufacturing throughput has been shown to improve by up to 50x with greater than 95% accuracy when compared to manual inspection. 

— Tim Juan, Senior Director, ADLINK Smart Factory Business Center

Agent Vi

“With OpenVINO™ toolkit the results have been impressive, enabling us to move from supporting two cameras to 20 with one developer in under three weeks. We will be able to fully scale our solutions to the edge with the right performance per dollar while leveraging Intel® Movidius™ Vision Processing Unit (VPU) and Intel® FPGA solutions.”

— Zvika Ashani, CTO and Co-founder, Agent Vi

Amazon Web Services (AWS)

“Intel’s wide range of deep learning acceleration silicon allows us to offer AWS Greengrass* to customers who require powerful yet cost-effective AI solutions at the edge. With [the] OpenVINO™ toolkit integrated [with] Greengrass devices, these customers are enabled to build cloud machine learning models that they can deploy at the edge to perform deep learning tasks in real time, which can be tailored to their specific performance needs. Our partnership with Intel in this space is essential to us providing customers a wide variety of cost-effective machine learning solutions.”

— Satyen Yadav, General Manager, IoT Ecosystem, AWS

AnyVision

“Because we serve customers with so many different needs, it’s important to quickly achieve the right balance of price and performance for each of our applications. Intel® DevCloud for the Edge lets us test multiple platforms in parallel. That’s a lot of time savings—and time is money—so it’s a no-brainer.”

— Eduard Vazquez, Research Technical Manager, AnyVision

CONEX

“Now through the Intel® AI: In Production program, CONEX will be able to rapidly go from our validated prototypes to actual end products. Thanks to this new path to production, we will be deploying an AI-enhanced, point-of-sale retail device to some of the largest cosmetic goods retailers as soon as this spring.”

— Nicolas Lorin, President, CONEX

CyberLink

"CyberLink is a general member of Intel® IoT Solutions Alliance, which aims to deliver the best AI at the edge solutions. We have optimized our key software applications with Intel® Distribution of OpenVino™ toolkit. We have integrated OpenVINO toolkit into a number of our creative software’s AI features such as styling videos into live paintings for video editing in PowerDirector* and photo deblurring in PhotoDirector*.

FaceMe*, our edge-based AI facial recognition SDK solution, is highly optimized for the OpenVINO toolkit. With OpenVINO toolkit acceleration enabled, the speed of FaceMe's facial recognition is improved by 500 percent (from 4.5 fps to 23 fps). Compared to running on Celeron® processor, FaceMe runs 17.4x faster on Intel® Movidius™ Myriad™ X VPU and saves 72 percent CPU usage.

Using these Intel® technologies with FaceMe not only enables use cases otherwise impractical or even impossible, but it can generate significant savings for our customers."

— Richard Carriere, Senior VP of Global Marketing and GM of Americas

DarwinAI

“We use machine learning to probe and understand a neural network on a very fundamental level, and then build up a very sophisticated mathematical understanding of the network during that process. We then use AI a second time to generate an entirely new family of neural networks that is considerably more compact than the original, as good as the original from a functional standpoint, but works a lot faster. That is the ‘generative’ in Generative Synthesis*. What’s more, the process is entirely complementary to the Intel® Distribution of OpenVINO™ toolkit, which does a superior job of optimizing mode performance on Intel® architecture. Specifically, by leveraging the Intel Distribution of OpenVINO toolkit, we saw up to 6.18x improvement in FP32 data types and 10.51x in INT8 running on 2nd Gen Intel® Xeon® Scalable processor-based platforms.”

— Sheldon Fernandez, CEO, DarwinAI

DeepSight

“Based on the Intel® Distribution of OpenVINO™ toolkit, we have developed a highly accurate, highly stable intelligent defect detection system for the tire industry. OpenVINO toolkit not only helps with significant inspection system performance improvement but also helps us achieve millisecond-level defect detection performance on a normal industrial computer with more than 99.9 percent accuracy. The solution helps the customer to cut labor costs by approximately USD 49,000 per production line per year, and the production efficiency has been improved greatly.” 

— Michael Li, Founder and CEO, DeepSight

DycodeX

"OpenVINO™ toolkit enables us to quickly develop a new smart camera solution under the SMARTernak* platform that implements visual intelligence at the edge. We’re able to finish an initial working product from two months to under two weeks by leveraging ready-to-use ML models, easy-to-use tooling, and Inference Engine API, while still achieving up to 10x inference throughput, even when it’s run on a single board computer at the edge that’s accelerated by Intel® Movidius™ Neural Compute Stick. The OpenVINO toolkit-powered smart camera offers 50 percent of the cost for our customers if compared to our wearable-based solution. For certain functionality (e.g., herd counting), the cost is reduced even further, as one camera is able to count the whole herd. OpenVINO toolkit certainly has tremendous value proposition for certain use cases that will greatly benefit our customers."

— Andri Yadi, CEO, DycodeX

European Space Agency (ESA)

"The market for remote sensing space-based application is fundamentally limited by the up and downlink bandwidth and the onboard compute capability on satellites. The work done by ESA shows how the compute capability on these platforms can be vastly increased by leveraging emerging COTS system-on-chip technologies. The orders of magnitude increase in processing power can then be applied to consuming data at the source rather than on the ground. This allows the deployment of value-added applications in space, which consume a tiny fraction of the downlink bandwidth that would be otherwise required. The solution, based on Intel® Movidius™ VPU, has the potential to revolutionize Earth Observation (EO) and other remote sensing applications, reducing the time and cost to deploy newly added value services to space by a great extent, at least 30 percent, compared with the current state. "

— Gianluca Furano, Data Systems Engineer 

— Massimiliano (Max) Pastena, Earth Observation Technology Management Engineer

GE Healthcare

“With the OpenVINO™ toolkit running on existing Intel® processors, GE Healthcare has achieved a 3.3x improvement in deep learning optimization, which enables early prioritization and escalation of critical conditions to ensure faster treatment for our patients. Intel® technologies will enable GE Healthcare to extend AI solutions across multiple imaging modalities to transform radiologist workflows and patient care.”

— Karley Yoder, VP and GM, Edison AI, GE Healthcare

GeoVision Inc.

"The implementation of OpenVINO™ toolkit has not only enhanced our processing capacity significantly as edge AI developers, but also surpassed for an optimal, ideal platform capable of fulfilling a wide range of AI applications. Statistically, it uplifted us to realize a performance gain of eight to tenfold after employing OpenVINO toolkit, making simultaneous, real-time video processing possible at greater capacities. Accompanied with detailed, intuitive documentations on model optimizer and inference engine API, the use of OpenVINO toolkit has also strengthened our development cycle of deep learning driven features.”

— Thomas Lee, Sales Engineer, GeoVision Inc.

Gorilla Technology

“At the Fengyuan Train Station, with the deployment of our OpenVINO™ toolkit-enabled system, we saw a 50 percent reduction in our emergency response time and a 70 percent decrease in Yellow Line crossings.”

—Taiwan Railway Bureau, MOTC

“We improved our Traffic Violation Management using an OpenVINO toolkit-optimized solution at one of our busiest intersections. Once online, we recorded a 50 percent drop in illegal left turn violations.”

— Kaohsiung Transportation Bureau

Gwell Medical 

Healthcare staff are busy and overworked, but the movement of supplies never stops. A single robot can do the work of up to three people and hold over 10x more weight than any single person, making hospital logistics a good job for robots. To meet the sensory requirements of  the hospital logistics distribution robots, Gwell Medical (formerly known as "MROBOT") Hospital Logistics Robot is powered by 7th generation Intel® Core i5/i7 processors (Intel® Core™ i5-7500 and i7-7700 processors), Intel® Movidius™ Myriad 2 VPUs, and optimized with Intel® Distribution of OpenVINO™ toolkit (allowing for a 5.8x increase in FPS throughput). The integration of these products and technologies provides a more energy-efficient and more rationally designed solution with unique, optimized processing capabilities for deep visual loads, which can meet the requirements of hospital logistics distribution for low-latency perception capabilities. These Intel based Hospital Logistic Robots have been applied successfully in more than 40 hospitals nationwide, including several Wuhan hospitals.”

— Harrison LV, CTO, Gwell Medical (formerly known as "MROBOT")

Hitachi Ltd.

“Because developers can quickly evaluate the performance of their applications in multiple edge computing systems by using Intel® DevCloud for the Edge, they can not only shorten the inspection time to go to market, they can also expect tremendous benefits in terms of investment and maintenance in verification equipment. We are confident that Intel DevCloud for the Edge will accelerate and streamline operations and create new value for more IoT businesses and for more customers.”

— Tomohiro Nagao, Senior Manager, Hitachi Ltd., Healthcare Business Unit

JLK Inspection

"By using the OpenVINO™ toolkit solution, we realize up to 90 percent faster speeds in the loading AI model, and that leads to undiminished performance on the CPU in the Intel® NUC system without GPU acceleration. As a result, there is a cost reduction of more than 50 percent, and the size is reduced by more than 70 percent. These improvements enhance a new value of mobility of medical AI solution, because the end-users can now easily implement AI modules into their Hospital Information System and maintain the system using Intel NUC with OpenVINO toolkit technology. Patients can be provided with a faster and more accurate diagnosis, which may lead to a better prognosis. Also, patients can be provided with the healthcare reports from the AI solution to easily know their health information."

— Dongmin Kim, Ph.D., CEO, JLK Inspection

MaxQ AI

“The original FDA-approved product release of Accipio Ix* took an average of 4.1 minutes to process a radiology exam. Thanks to the optimizations with Intel® Distribution of OpenVINO™ toolkit, the average processing time for the first 4,000 exams at Capital Health has been 1.4 minutes.” 

— Steve Kohlmyer, VP Research and Clinical Collaborations, MaxQ AI

NCTech Ltd.

“The framework independent OpenVINO™ toolkit inference engine allowed NCTech to build an advanced data privacy engine with similar accuracy as seen in Google Street View*. Data privacy is to remove identifiable personal information (such as people's faces or car number plates) from street-level images. We use AI to detect people's faces and car number plates in images, and then we blur them so they are not identifiable. To achieve the best accuracy, we utilize different AI frameworks for training and OpenVino toolkit as our inference engine.

This improvement in speed allows us to process and stitch data sets captured per one mile in 30 minutes instead of the two hours it took for the original system, reducing costs by up to 75 percent."

— Cameron Ure, CEO and Co-founder, NCTech Ltd.

QNAP

"We at QNAP, along with one of the leading hospitals in Taiwan, have jointly worked to bring in an innovative assistive healthcare solution in the area of skin disease diagnosis. Based on the strong technical foundation of Intel® Distribution of OpenVINO™ toolkit, this is a non-invasive, AI-based solution that helps to identify five different types of skin diseases that were difficult to diagnose earlier. Intel Distribution of OpenVINO toolkit has played an instrumental role in achieving the expected performance and accuracy of the solution." 

— Y.T. Lee, Vice President, QNAP

QuEST Global

“QuEST used Intel® Distribution of OpenVINO™ toolkit for our solution to helping radiologists identify lung nodules through Computer Aided Detection (CAD) models. We obtained impressive results while we employed model inference on Intel® Edge devices without compromise in performance. We achieved 33x performance improvement in running the OpenVINO toolkit model on Intel® Core™ processor-based edge devices when compared to the unoptimized baseline model. Intel® architecture with OpenVINO toolkit offers an affordable solution when trying to balance accuracy and complexity of the model and real time performance. This means we don’t have to opt for new, expensive, hardware-based deployment around healthcare institutions, resulting in a strong TCO advantage."

— Dr. Rubayat Mahmud, Director of Sales and Business Development, QuEST Global

Rosmart Technology Co., Ltd.

"Defect detection is the key process in electronic component manufacturing. However, manual inspection performed by operators cannot meet the quality and efficiency requirements, otherwise labor cost will be increased quickly. We design and implement automatic defects detection machine based on Deep Learning algorithm that is accelerated by Intel® Distribution of OpenVINO™ toolkit. We have seen great improvement of the inspection accuracy from 87% to 98%, and cuts labor cost by nearly RMB 500,000 (an equivalent of 70,500 US dollar) per production line per year. In addition, Intel® Distribution of OpenVINO™ toolkit not only helps us to deploy AI model on Windows10 platform easily, but also to cut computing cost by nearly RMB 10,000 (an equivalent of 1400 US dollar) per equipment comparing to the original GPU inference solution."

— Johnson Li, Co-Founder and VP of Guangdong Rosmart Technology Co., Ltd.

Topaz Labs

“Topaz Labs worked with Intel on our latest release of our Gigapixel AI* for Photo. Gigapixel AI uses AI/ML and deep-learning methods to enlarge images and add detail. With OpenVINO™ toolkit implementation, we saw approximately an 8.73x performance increase of photo upscaling up to 400 percent while preserving image quality using AI/ML. The model optimizations through OpenVINO toolkit and targeting Gen 11 graphics resulted in a 2.93x performance benefit on Ice Lake compared to the competition and a 1.74x Gen over Gen!”

— Eric Yang, CEO, Topaz Labs

“Sharpen AI* is the first sharpening and shake reduction software that can tell the difference between real detail and noise. Our customers have seen a greater than 7x improvement in sharpening, just under an 11x increase when stabilizing, and a 10.5x improvement when focusing since the implementation of OpenVINO toolkit into our Sharpen AI application.”

— Russell Tarpley, VP of Products, Topaz Labs 

“DeNoise AI* uses artificial intelligence and machine learning, which together allow the elimination of noise and the recovery of crisp detail in photos images with the first AI powered noise reduction tool. Gen 11 GPU performance on the 10^th Generation Intel® Core™ processor is 1.86x better than KBL-R and 3.57x better than Ryzen 3700U* for denoising images. Optimization of DeNoise AI using OpenVINO toolkit produced over a 7x performance improvement compared to the previous version!”

— Albert Yang, CTO, Topaz Labs

Touch Cloud, Inc.

"OpenVINO™ toolkit improves inference performance up to 10 times faster, reduces CPU loading, and enables an edge device to support more channels."

— Simon Lee, Co-founder and CEO, Touch Cloud, Inc.

Videonetics

Intel® Distribution of OpenVINO™ toolkit and DL streamer helped us to achieve greater performance using the same range of processors with significant reduction of development time. By leveraging the decoding and deep learning computing capability of Intel’s built-in GPU on the processors, our deep learning engines are now distributed at various layers of our fog computing framework in video-based IoT solutions. Intel® Distribution of OpenVINO™ toolkit and DL streamer helped us to quickly come up with a new generation of applications in intelligent traffic management and intelligent video analytics, and the smart urban analytics domain with a very low cost of ownership to our customers, thus enabling them to deploy these solutions at a much larger scale.

— Tuhin Bose, VP and CTO, Videonetics Technology Pvt. Limited

Vispera Information Technologies

As a member of the Intel® AI: In Production Program, we use Intel® DevCloud for the Edge and Intel® Distribution of OpenVINO™ toolkit extensively, where CPU-run engines that are optimized with the latter are currently in production. So far, We have used DevCloud for the Edge primarily for benchmarking purposes, which enabled us to determine the optimal edge hardware configurations, algorithmic choices and engine parameters, production-level load estimations and scalability assessment under different production scenarios. With our solution, the results obtained from the exercise in DevCloud for the Edge show that computing deep neural networks on CPU has become on par with GPU in terms of performance, and it is constantly improving with newer tools like int8 quantization, making advanced edge inference solutions like ours feasible with CPU. Using OpenVINO toolkit, the per image inference time speeds up by 10x as compared to unoptimized CPU execution, which allows us to run our solution as a real-time IoT application operating on Intel® Xeon™ processors at multiple stores.

— Erdem Yoruk, Chief Scientist, Vispera Information Technologies

VSBLTY

"The implementation of OpenVINO™ toolkit in the VSBLTY product suite for true edge-based computer vision in retail environments was a god-send at a time when we really needed it. We have customer implementations all over the world. We built the fault tolerance for low bandwidth and intermittent internet connectivity that nicely handles the audience measurement analytics we are doing. OpenVINO toolkit provided the software plumbing we so desperately needed to run our computer vision algorithms on the edge in real time without the need for any bandwidth, let alone network connectivity. VSBLTY gained huge performance increases, in many cases 4 to 5x as fast, in the processing of our computer vision machine learning models.” 

— Tim Huckaby, CTO, VSBLTY

Wahtari

"Due to our modular AI-Training Platform, we are using many different AI frameworks, where ONNX* simplifies the deployment processes of trained models, since they adhere to one shared standard. OpenVINO™ toolkit is backed by a strong open-source community, great documentation, a deep integration into OpenCV, and an easier to use API compared to competitors like TensorRT*, CuDNN*, etc. It also makes it possible to switch between different hardware for execution, like CPU, GPU, FPGAs, and Intel® Movidius™ technology, on-the-fly." 

— Sebastian Borchers, Co-founder, Wahtari

WonderStore

"As an Intel® AI: In Production Program member, WonderStore is dedicated to Intel® Distribution of OpenVINO™ toolkit, which accelerates CV models on Intel-based hardware. Because WonderStore creates specific CV models for specific customers, we need OpenVINO to be able to run many models smoothly, using affordable hardware. By leveraging a huge repository of publicly-available models for the CV developer community offered by the toolkit, WonderStore can provide in-store analytics and detailed shopper profiles: not just gender and age, but also which clothing and accessory brands they wear. " 

— Reinier van Kleij, PhD., Partner, Business & Technology, CTO, WonderStore 

"WonderStore uses Intel® DevCloud for the Edge to train models with data sets of over 30,000 pictures in half the time vs. on-premise servers.  Intel® DevCloud for the Edge allows us to choose our hardware configuration to optimize training, enabling us to create vertical CV models for each customer's needs." 

— Reinier van Kleij, CTO, WonderStore