search

nnstreamer / nnstreamer

:twisted_rightwards_arrows: Neural Network (NN) Streamer, Stream Processing Paradigm for Neural Network Apps/Devices.
https://nnstreamer.ai
GNU Lesser General Public License v2.1
683 stars 171 forks source link

Support for Raspberry Pi AI Kit? #4483

Closed bveldhoen closed 4 days ago

bveldhoen commented 2 weeks ago

Add support for Raspberry Pi AI Kit / Hailo-8? See here: https://news.itsfoss.com/raspberry-pi-ai-kit/

It uses the Hailo 8 Edge AI processor: https://hailo.ai/products/ai-accelerators/hailo-8l-m-2-ai-acceleration-module-for-ai-light-applications/?ref=news.itsfoss.com#hailo8lm2-overview

I assume this isn't supported yet (Pls. correct me if I'm wrong!). Are there any plans to add support for this?

taos-ci commented 2 weeks ago

:octocat: cibot: Thank you for posting issue #4483. The person in charge will reply soon.

myungjoo commented 2 weeks ago

If this requires a new user library or runtime (Hailo RT?) to run, yes, you need a new tensor_filter subplugin for Hailo.

I do not have a plan to support this.

Anyone who's interested in this may implement (by adding a tensor_filter_subplugin C++ subclass) and upstream.

bveldhoen commented 6 days ago

There is a repository with a number of gstreamer components here: https://github.com/hailo-ai/tappas

Do you see a good way to integrate with these elements?

myungjoo commented 5 days ago

If they are GStreamer elements, they can be integrated at run-time via GStreamer as long as they are compiled for the same architecture and OS. You just need to install all components in one system.

Are there any issues against uring them in a single system?

It appears that TAPPAS "filter" accepts RGB video stream as inputs and outputs. As long as you take care of stream data types (GST CAPS, usually handled by tensor_converter, and tensor_decoder in nnstreamer), it should be compatible with nnstreamer.

bveldhoen commented 4 days ago

Thank you! I might give it a try later. Closing for now.