Basically, this repo is driven by AI models. So the author won't add as many nodes as possible to cover all ONNX's ops. It's RECOMMENDED that you contribute in these situations:
- Some bugs need to be fixed;
- New public models support;
- Code optimization;
- New features;
- Complete TODOs.
It's NOT RECOMMENDED that you contribute if your code can't be tested by public models. You should keep it in your private branch,
even if this op is a popular one. Because there will be a potential risk of merging untested codes.
There are some suggestions to avoid sensitive data leaking to the public:
- Use onnx.helper to build a similar ONNX model;
- Use onnx_tool to save a model without any weight;
- Use onnx_tool to get a subgraph model with your interested part only;
- Introducing your changes;
- How to test your changes: how to download the ONNX model or how to create a same model.