Graph annotation based on attribute values

[lutzroeder]

Scenario:

UI feature to augment the graph with a visualization of a selected attribute, metadata or metrics values.

The selected attribute should both be shows as part of each node and if float turned into a border or background style.

Questions:

• Should the feature get activated from property value by selecting a metrics of by filtering (Find) based on criteria?
• What visualization is most intuitive?

[SStarver]

Mark, it would be fantastic to support this feature. Recently, I'm looking for a method to fill accuracy analysing result into each onnx model tensor, I think this feature is useful to satisfy my demandings.

[sheny1xuan]

Looking forward to this feature.

[xieofxie]

Looking forward to this! For example, I want to debug QDQ to know which node significantly goes wrong, I could use this https://github.com/microsoft/onnxruntime-inference-examples/blob/main/quantization/image_classification/cpu/run_qdq_debug.py to generate a json file with name and accuracy, then show it side by side with graph in the netron