Low-Power Image Identification at 0.35 Watts
The Grove Vision AI Board, a new innovation in edge AI projects, is designed to deliver impressive performance with minimal resources. Developed by Jaryd, this board is set to revolutionise the way we approach AI in everyday devices.
One of the key features of the Grove Vision AI Board is its ability to run alternative image recognition models, though their quality may vary. The board's standout model is YOLO8, which, despite recognising only one object at a reduced resolution of 192×192 and being quantized down to INT8, delivers a performance of 20-30 frames per second (fps) with good accuracy.
The Grove Vision AI Board's power consumption is another area where it shines. It operates on just 0.35 Watts of power, making it more energy-efficient than a single key on a typical RGB-backlit keyboard.
The board's capabilities are not limited by its power usage, however. It can be connected to a camera module and a host device, which can range from another microcontroller to a more powerful computer. When signalled by the host device, the Grove board takes a picture, performs image recognition, and sends the results back to the host device.
Speech recognition can also be run on more limited devices with edge AI projects, making the Grove Vision AI Board a versatile tool for various applications. Moreover, it is possible to train one's own image recognition model for the Grove Vision AI Board, providing users with the freedom to customise their AI solutions.
At the heart of the Grove Vision AI Board is the WiseEye processor, which combines two ARM Cortex M55 CPUs and an Ethos U55 NPU for AI acceleration. This powerful combination ensures that the board can handle complex AI tasks while maintaining its low power consumption.
While some people are attempting to make image recognition less effective in edge AI projects, the Grove Vision AI Board is designed to address the expense and backlash associated with the massive power consumption of AI models. It offers decent image recognition results with less power consumption and hardware, at the cost of some ability and accuracy.
In the example provided by Jaryd, a UART was used instead of the I2C library to signal the board. This flexibility in signalling methods further enhances the Grove Vision AI Board's versatility.
In conclusion, the Grove Vision AI Board is a promising development in edge AI, offering a balance between performance, power consumption, and hardware requirements. Its ability to run alternative image recognition models, its energy efficiency, and its versatility make it a valuable tool for a wide range of AI applications.
