How to render custom ONNX annotations

Hi! First of all, thank you for making Netron and sharing it with the community. We use Netron in our quantized neural network compiler FINN (https://github.com/Xilinx/finn) to visualize ONNX, including as part of Jupyter notebook visualizations and it’s been immensely useful.

I would like to extend Netron to render some ONNX annotations but I don’t know where to start, so I thought I’d ask for some help here first. In our project we work with few-bit integer tensors and we use custom tensor annotations in ONNX to represent them, by using the ONNX quantization_annotation field to carry strings that indicate the datatype. I want to add the possibility of visualizing the tensor quantization annotation next to (or instead of) the tensor shapes that are printed, something like this:

How would I go about doing something like this? Any tips would be highly appreciated!

[lutzroeder]

This would be somewhat in line with how quantization information is rendered for other formats:

[maltanar]

Fantastic, thank you!