New plugin for ONNXRuntime

[MyungJoo Ham]

On Fri, 10 Jun 2022 at 12:15, pranav sharma via gstreamer-devel <
gstreamer-devel at lists.freedesktop.org> wrote:

> Hi, I'm from Microsoft and we're planning to build a plugin for ONNX
> Runtime. ONNX Runtime is a very popular open source AI framework. You can
> read all about it here: https://onnxruntime.ai/. I see that a plugin for
> ONNX has already been checked in here
> https://gitlab.freedesktop.org/gstreamer/gstreamer/-/tree/main/subprojects/gst-plugins-bad/ext/onnx.
> However, it's marked "bad". I've a few questions around this.
> 1. When a plugin is first checked in, does it always get merged into the
> "bad" folder?
> 2. What's the process of getting the plugin merged into the good category?
> 3. What type of information is required while checking in the plugin to
> classify it as a good plugin?
>
> Thanks,
> Pranav.
>

Hi,

This might a bit off-topic to your questions; however another approach might be helpful.


If you want to write GStreamer pipelines for arbitrary machine learning applications,
nnstreamer (a set of GStreamer plugins for neural network apps at github.com/nnstreamer),
can be another approach for you.


For each neural net framework or hardware accelerator, it accepts a plugin for its
inference plugin (tensor-filter).
The "upstreamed" plugins for inference plugin are located at:

https://github.com/nnstreamer/nnstreamer/tree/main/ext/nnstreamer/tensor_filter



Cheers,
MyungJoo