June 5, 2024
news

Raspberry Pi now has AI available on its platform that is specifically tailored for low-power edge applications. The new Raspberry Pi AI Kit is available now for $70 and helps developers experiment with and implement neural networks, artificial intelligence, and machine learning on the Raspberry Pi 5.
The AI Kit was developed in collaboration with Hailo, and the partners say the kit is designed to provide an accessible way to integrate local, high-performance, power-efficient inference into a variety of applications, enabling the rapid building of complex AI vision applications that run in real-time with low latency and low power requirements. Neural networks for object detection, semantic and instance segmentation, pose estimation, and facial landmark detection can all reportedly run entirely on the built-in Hailo-8L coprocessor.
The kit delivers 13 TOPS of inference performance and has a single-lane PCIe 3.0 connection that runs at 8Gbps. It is of course fully integrated with the Raspberry Pi imaging software subsystem and is compatible with first- or third-party cameras, according to the documentation. If you don't want to build your own neural net, Hailo offers a variety of pre-trained neural network models that are ready to deploy.
The focus of the software has been to enable AI vision applications on the platform. As such, the software is compatible with first and third-party cameras, and the rpicam-apps suite of camera applications includes post-processing templates for integrating neural network inference running in real-time in the camera pipeline. Leveraging the pre-installed Hailo Tappas post-processing library, users can create AI apps with just a few hundred lines of C++ code.
Of course, there are hundreds of possible applications, but Raspberry Pi has created a few previews, check them out here.
The partners did not limit users to just rpicam-apps or Picamera2, but also created an API integrated into the GStreamer framework and native Python or C/C++ applications for off-camera use cases, such as running inference on recorded video files.
For more information, see Hailo's product webpage for the AI Kit documentation and full technical specifications for the Hailo-8L AI accelerator module.
