I have got a Raspberry Pi 5 and the AI Kit. My understanding is this: To create new models I have to use the AI Software on an Ubuntu computer with around 100 GB disk space and 32GB memory. This cannot be done on the Raspbery Pi 5.
My question is this: Can someone give step by step instructions (including a sample model with its name) on how to use an existing model from the Model Zoo on Raspberry Pi 5 Hailo hardware and see the results on the RPI monitor? These basic but fundamental instructions are missing from everywhere I looked on the Internet. There seems to be too many sites on AI but none of them teaching anything!
This information is very helpful. Is there any info about how many NPU cores there are on the 8L, or how CPU/memory intensive running models with it is assuming that most operations are mapped to it?
I managed to run the demos at the “Github-haolo-ai/hailo-rpi5-examples” you mentioned below. But, I could not find any demos to run under the link you gave as”Github-hailo Hailo-Application-Code-Examples” Application Code Examples. How do I run them using the Hailort on RPI?
The Hailo accelerators do not have NPU cores. They have 3 types of resources (compute, memory and control). More like an FPGA. The Hailo Dataflow Compiler will map each network differently. The mapping for each layer depends on the layers requirement AND the entire network. The same layer can use different amounts of compute resources depending on what other layers need to be implemented. This way we can get the best implementation for each network making as much use of the chip as possible.
This depends on each model, the number of sources and required framerates. Each model has different requirements for the pre- and post-processing.
The Hailo Application Code Examples cover many different use cases. They are not designed to run out of the box on any platform but show you some possibilities with different frameworks and languages.