AI Models
Kneron produces edge AI models that are built upon mainstream AI frameworks in line with our focus to balance performance and power consumption. Our software mirrors our hardware in that we build for edge use cases that prioritize TOPS per Watt.
Kneron focuses on the development of lightweight AI models, and possesses patented quantization technology, which renders the model smaller yet with the same precision.
In July 2019, the National Institute of Standards and Technology (NIST) released the latest Face Recognition Vendor Test (FRVT) competition report. Among all proposed models within 64MB, NIST ranked the Kneron-003 modelfirst.
In 2020, Kneron's low-cost binocular face recognition algorithm passed the UnionPay BCTC certification, and been applied to multiple brands of smart door locks.In addition, the company has developed various algorithms aiming for object and vehicle recognitions, and has cooperated with numerous worldwide brand name customers in fields of ADAS and AIoT.
In July 2019, the National Institute of Standards and Technology (NIST) released the latest Face Recognition Vendor Test (FRVT) competition report. Among all proposed models within 64MB, NIST ranked the Kneron-003 modelfirst.
In 2020, Kneron's low-cost binocular face recognition algorithm passed the UnionPay BCTC certification, and been applied to multiple brands of smart door locks.In addition, the company has developed various algorithms aiming for object and vehicle recognitions, and has cooperated with numerous worldwide brand name customers in fields of ADAS and AIoT.
IMAGE RECOGNITION MODELS
In-line with our hardware offerings, Kneron's AI models and algorithms maintain low latency, high accuracy, and strong performance. They balance power, model size, and complexity. Whether in drones flown off the grid, in robotics that run on battery power, or smaller form factors like smart doorbells, door locks, and AI cams that need edge AI models that are under 50MB and operate up to 6 months, Kneron's AI models fit.
Kneron's lightweight image recognition algorithms can recognize human faces, bodies, gestures, objects, and scenes on-device, and power applications such as secure payments, door security, facilities management, retail management, and traffic management.
The key to effective edge AI computing is balancing size, performance, and power usage. That said, Kneron's facial recognition algorithm (Kneron-003) scored the best aggregate performance rating among all lightweight models under 100MB in the U.S. National Institute of Standards and Technology (NIST) 2019 Face Recognition Vendor Test (FRVT).
The key to effective edge AI computing is balancing size, performance, and power usage. That said, Kneron's facial recognition algorithm (Kneron-003) scored the best aggregate performance rating among all lightweight models under 100MB in the U.S. National Institute of Standards and Technology (NIST) 2019 Face Recognition Vendor Test (FRVT).