Implement some ONNX exports, needed for SSD

[RuRo]

Description

This is a WIP pull request for implementing exports for some ONNX operators, which are needed to successfully export the SSD model from gluocv. The only 2 operators, which are not yet implemented are _contrib_box_decode, which can be sidestepped by using SSD(minimal_opset=True) and _contrib_box_nms, which could be potentially implemented with the NonMaxSuppression ONNX operator, but this is a big enough undertaking to be its own separate PR.

This PR assumes that #18054 is merged (ruro/onnx_version_bump), since it uses some ONNX operators only available in v1.5.0. (The first commit is a squashed version of that PR and will be dropped after it gets merged into master)

If you actually want to try exporting SSD, you should also merge #17827 (QueensGambit/onnx_export_slice_fix or ruro/onnx_export_slice_fix).

Changes

• eagerly check onnx conversion availability in verbose mode
  The export_model method now warns about all nodes, that don't have a registered conversion before trying to export the model, if verbose=True. This is a simple quality of life change, which helps potential contributors to estimate, how many layers are missing until a model can be exported.
• implement _arange op
  (doesn't support the repeat attribute for now)
• implement ones_like and zeros_like ops
• implement _lesser_scalar, _greater_scalar and _equal_scalar ops
  also fixes the issue, where the scalar_op_helper would fail, if the input is a non-1D initializer
• implement slice_like op
  (doesn't support the implicit axes variation right now)
• implement where op

[mxnet-bot]

Hey @RuRo , Thanks for submitting the PR
All tests are already queued to run once. If tests fail, you can trigger one or more tests again with the following commands:

• To trigger all jobs: @mxnet-bot run ci [all]
• To trigger specific jobs: @mxnet-bot run ci [job1, job2]

---

CI supported jobs: [windows-gpu, centos-gpu, windows-cpu, miscellaneous, clang, unix-gpu, centos-cpu, website, sanity, edge, unix-cpu]

---

Note:
Only following 3 categories can trigger CI :PR Author, MXNet Committer, Jenkins Admin.
All CI tests must pass before the PR can be merged.

[mxnet-bot]

Jenkins CI successfully triggered : [unix-gpu]

[VisionZQ]

Hi,
Thanks for you kindness.
We have converted the mxnet to onnx 1.5, and the 'slice_like' is work.
But, When I from onnx convert to TensorRT6.0, the program crash. can you give me any suggestion?
While parsing node number 205 [Gather -> "yolov30__mulscalar0__shape__gathered"]: ERROR: /home/jakin/songzq/onnx-tensorrt/onnx2trt_utils.hpp:412 In function convert_axis: [8] Assertion failed: axis >= 0 && axis < nbDims
Thanks a lot.

[RuRo]

@VisionZQ This PR currently has WIP status (it's not ready) and it's main purpose was to implement the needed ops to convert SSD models from mxnet to ONNX. There's been a lot of breakage recently with onnx2mx, so some further work may be required. I am not sure, I can help you with converting yolov3. I don't currently have the free time to even finish this PR.

I would recommend, that you open an issue, describing, which layers needed for yolov3 don't have a conversion in mxnet 1.6.0 and which conversion functions are currently missing in mxnet 2.0.0, but I wouldn't get your hopes up, that somebody will implement the conversions for you. Afaik, most mxnet contributors are currently very busy with mxnet 2 API changes.

[chinakook]

It's a great job!

[Aktcob]

Really good job！

When I convert yolov3 model in gluoncv to ONNX, I'm faced with the same problem.