Add more advanced smoke test for project Nova and validation workflows

[atalman]

Move vision smoke test residing in builder: https://github.com/pytorch/builder/blob/main/test/smoke_test/smoke_test.py
So that we can support these from vision repo and share them with project Nova.

[osalpekar]

LGTM on my end once everything is green, thanks!

[datumbox]

LGTM, thanks!

FYI any of the Object Detection models would provide a better coverage than Classification because it uses custom ops like RoI Align. Usually it's better to send actual data through the layer to check that it actually works. At least that's what we found out few months back when the M1 changes landed and we faced an issue with the dispatcher.

failing test

[datumbox]

@atalman I just noticed that one of the tests fails with:

ERROR conda.cli.main_run:execute(41): `conda run python3 pytorch/vision/test/smoke_test.py` failed. (See above for error)
/Users/runner/work/_temp/conda_environment_3634174756/lib/python3.10/site-packages/torchvision/io/image.py:13: UserWarning: Failed to load image Python extension: dlopen(/Users/runner/work/_temp/conda_environment_3634174756/lib/python3.10/site-packages/torchvision/image.so, 0x0006): Symbol not found: (__ZN2at4_ops19empty_memory_format4callEN3c108ArrayRefIxEENS2_8optionalINS2_10ScalarTypeEEENS5_INS2_6LayoutEEENS5_INS2_6DeviceEEENS5_IbEENS5_INS2_12MemoryFormatEEE)
  Referenced from: '/Users/runner/work/_temp/conda_environment_3634174756/lib/python3.10/site-packages/torchvision/image.so'
torchvision: 0.13.1a0
  Expected in: '/Users/runner/work/_temp/conda_environment_3634174756/lib/python3.10/site-packages/torch/lib/libtorch_cpu.dylib'

  warn(f"Failed to load image Python extension: {e}")
Traceback (most recent call last):
  File "/Users/runner/work/_temp/conda_environment_3634174756/lib/python3.10/site-packages/torch/_ops.py", line 563, in __getattr__
    op, overload_names = torch._C._jit_get_operation(qualified_op_name)
RuntimeError: No such operator image::decode_png

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/Users/runner/work/vision/vision/pytorch/vision/test/smoke_test.py", line 63, in <module>
    main()
  File "/Users/runner/work/vision/vision/pytorch/vision/test/smoke_test.py", line 57, in main
    smoke_test_torchvision()
  File "/Users/runner/work/vision/vision/pytorch/vision/test/smoke_test.py", line 17, in smoke_test_torchvision
    all(x is not None for x in [torch.ops.image.decode_png, torch.ops.torchvision.roi_align]),
  File "/Users/runner/work/_temp/conda_environment_[363](https://github.com/pytorch/vision/actions/runs/3634174756/jobs/6132002165#step:11:364)4174756/lib/python3.10/site-packages/torch/_ops.py", line 567, in __getattr__
    raise AttributeError(
AttributeError: '_OpNamespace' 'image' object has no attribute 'decode_png'

It seems the operator is not registred properly?

[atalman]

LGTM, thanks!

FYI any of the Object Detection models would provide a better coverage than Classification because it uses custom ops like RoI Align. Usually it's better to send actual data through the layer to check that it actually works. At least that's what we found out few months back when the M1 changes landed and we faced an issue with the dispatcher.

@datumbox could you please point me to an example of this? I can try to add additional test using Object Detection model, however probably will open new PR for this

[datumbox]

Tests passing, approve.

[github-actions]

Hey @atalman!

You merged this PR, but no labels were added. The list of valid labels is available at https://github.com/pytorch/vision/blob/main/.github/process_commit.py

[datumbox]

@atalman concerning the Object Detection example, we have one in our documentation here. It's extremely similar to what you have in this PR.