AI startup Kneron launches NPUs for on-device facial and gesture recognition
Artificial intelligence startup Kneron has announced the low power Kneron NPU IP Series with three neural processing units (NPUs) for drones, smartphones, and IoT devices.
The company was founded in 2015, is headquartered in San Diego and backed by Qualcomm and Alibaba. Its customers include a Chinese home appliance company which will ship some 50 million devices embedded with Kneron AI solutions within the next year, according to Qualcomm Ventures.
The Kneron NPU IP Series runs at less than 0.5W, and allows deep learning networks like ResNet and YOLO to run on offline devices, and supports various convolutional neural network models and deep learning frameworks, the company says.
The KDP 300 is billed as an ultra-low power processor designed for facial recognition in smartphones. It consumes less than 5mW, and supports faster and more accurate 3D facial recognition with a convolutional neural network processing 5 frames per second at 20MHz, according to the announcement.
The KDP 500 enables deep learning, along with real-time recognition and analysis for multiple faces and for hand and body gestures. It is meant for smart home and smart surveillance applications, and consumes 100mW to process up to 152 GOPS (groups of pictures) at 500MHz.
The KDP 700 is in development, and supports advanced AI applications and deep learning inference for high-end smartphones, robots, drones, and smart surveillance devices.
MIT researchers recently announced the development of a new low power chip for running neural networks locally on mobile or IoT devices.