torchvision Resize export doesn't support antialias=True, ::_upsample_bicubic2d_aa to ONNX opset version 16 is not supported
Open
Nobody has claimed this yet.
module: onnx
- Dominant language
- Python
- Stars
- 17.9k
- Forks
- 7.3k
- Avg merge
- 1d 15h
- Merged PRs (30d)
- 13
Description
🐛 Describe the bug
'''python
resize = torchvision.transforms.Resize(size=(224, 224), interpolation=torchvision.transforms.InterpolationMode.BICUBIC, max_size=None, antialias=True)
dummy_input = torch.randn(1,3,2000,2000)
torch.onnx.export(resize, dummy_input, preprocess_model_path, input_names=['image'], output_names=['processed_image'], dynamic_axes={
# dict value: manually named axes
"image": {2:'h', 3:'w'},
})
'''
error "torch.onnx.symbolic_registry.UnsupportedOperatorError: Exporting the operator ::_upsample_bicubic2d_aa to ONNX opset version 16 is not supported."
Versions
ONNX opset version 16 is not supported
cc @neginraoof @BowenBao
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by reproducing the torchvision.transforms.Resize example with BICUBIC interpolation and antialias=True, then trace the torch.onnx.export path for the ::_upsample_bicubic2d_aa operator at ONNX opset 16. Done means this configuration no longer fails with the reported UnsupportedOperatorError.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- computer-vision
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100