Discover How Image Recognition Achieves Powerful Results with Only 0.35 Watts of Energy

Developing AI models can be very costly, and one big reason is their high energy use. But if you’re open to making some trade-offs in accuracy and speed, you can still get good results with less powerful hardware. For instance, the Grove Vision AI board packs a punch while only consuming 0.35 Watts. That’s a fraction of what many devices use!

The secret behind this efficiency is the WiseEye processor. It features two ARM Cortex M55 CPUs alongside an Ethos U55 NPU, which helps with AI tasks. This board connects to a camera module and a more powerful device, like a microcontroller or a computer. When the host device sends a signal, the Grove board snaps a picture, performs image recognition, and sends the results back. The setup simplifies communication using I2C, though in one example, [Jaryd] opted for UART.

To keep power usage low, the image recognition model has some constraints. It can run YOLOv8 but can only identify one object at a time, operate at a resolution of 192×192, and needs to be quantized to INT8. Despite these limits, it achieves impressive performance—20-30 frames per second with good accuracy. Interestingly, as [Jaryd] highlights, it uses less power than a single key on a typical RGB keyboard. If you’re looking for different models, there are a variety available, but quality can vary. And if you can’t find what you need, training your own model is always an option.

Edge AI projects like this focus on maximizing performance with limited resources. If your needs are more lenient, you can even run speech recognition on simpler devices. Nonetheless, it’s notable that some people actively try to reduce the effectiveness of image recognition.

As AI continues to evolve, the emphasis on energy-efficient models is growing. According to recent studies, AI-related energy consumption could triple by 2025 if we don’t pivot towards greener technologies. Experts suggest that innovations like edge AI can significantly reduce this footprint. Balancing performance and power will be crucial as we move forward in the field.

In summary, while AI development can be intensive, there are smart ways to get good results without heavy power use. The Grove Vision AI board is a prime example of this trend towards efficiency in tech.

Source link