[Feature Request] Classify Contributed Ops

[GeorgeS2019]

Describe the feature request

Currently there is no categorization of contributed Ops into e.g. NLP, Vision, and Audio

All are under ONE same namespace: com.microsoft

The suggestion is to split these Ops into the following categories

e.g.

• [ ] com.microsoft.nlp
• [ ] com.microsoft.vision
• [ ] com.microsoft.audio

For those Ops that are non-specific to any one of the above categories, they stay with

• [ ] com.microsoft

Describe scenario use case

To democratize the use of these ops, especially for new users, for users interested on e.g. NLP, they could look into the following namespeace

nlp

com.microsoft.nlp

• com.microsoft.nlp.BeamSearch
• com.microsoft.nlp.GreedySearch

vision

com.microsoft.vision

• com.microsoft.vision.CropAndResize

[GeorgeS2019]

This proposed categorization will facilitate discussion and issue project management of existing feature requests/discussions:

• [ ] [Documentation]: Overview of C# Workflow for consuming "Augmented" Onnx model with Custom Operators
• [ ] C# pre and postprocessing with opencvsharp for onnxruntime
• [X] OpenCV Image Decoder & SuperResolution CustomOps

[tianleiwu]

Our operator need to be backward compatible so it is not feasible to change domain without breaking existing application.

I think adding some tags in operator documentation might serve the purpose.

[GeorgeS2019]

flowchart TD
    image-->|com.microsoft.vision\nops| onnx\nmodel-->|com.microsoft.vision\nops| output\nimage

flowchart TD
    text-->|com.microsoft.nlp\nops| onnx\nmodel-->|com.microsoft.nlp\nops| output\ntext

[GeorgeS2019]

com.microsoft.nlp

use case 1 : [Performance] how I can stop or truncate the generation in onnx gpt-2 model with the beam search option