ONNX 1.3 Release

Release Note [Preliminary]

Snapshot release of master commit onnx/onnx@bae6333

• ONNXIFI 1.0
  • ONNX native backend interface
• Operator Set 8
  • Control Flow Operators graduated from experimental
    • Loop, If and Scan
  • Added new operator Expand
  • Updated operators Max, Min, Mean and Sum to support broadcasting
  • Support output indices in operator MaxPool
  • Varies documentation improvements
• Introduced Function concept for representing composed operators [experimental]
  • MeanVarianceNormalization is our first example of using this feature
• Enhanced shape inference
  • Support shape inference for Reshape operator with constant new shape
• More ONNX optimization passes
  • Available passes are here
• More operator backend tests
  • New test coverage stat page is added
• Opset Version Converter
  • Supported operators include: Add, Mul, Gemm, Relu, BatchNorm, Concat, Reshape, Sum, MaxPool, AveragePool, Dropout
  • All models in model zoo are covered, except tiny-yolo-v2 (PRelu needs adapter, WIP)
• Quantization coming soon
  • We are currently working with the community to collect more feedback and finalize. We expect this to happen quickly and will be released as quickly as possible and out of cycle if needed.

---

Shall we kick off the effort around the 1.3 release? My initial thoughts:

• We target a clean release from master (not a cherry pick)
• Keep RNN/CF as experimental
• Converge and merge quantization (if possible)

Thoughts?

cc @Maratyszcza @prasanthpul @linkerzhang @lupesko @houseroad @bddppq

[Maratyszcza]

We plan to release ONNXIFI 1.0 together with ONNX 1.3

[postrational]

It would be great if packaging for 1.3 addressed the segmentation fault issues users have been seeing. Here are some of the reports: https://github.com/onnx/onnx/issues/963, https://github.com/onnx/onnx/issues/1056, https://github.com/onnx/onnx/issues/1079, https://github.com/onnx/onnx/issues/1187.