Closed ghost closed 3 years ago
Unfortunately, at the moment ONNX does not support the "mv" operator, so exporting PyTorch models seems to fail. We'll have to wait and see for a while.
MODEL_URL = "https://www.dropbox.com/s/starqc9qd2e1lg1/ArtLine_650.pkl?dl=1"
urllib.request.urlretrieve(MODEL_URL, "ArtLine_650.pkl")
path = Path(".")
learn=load_learner(path, 'ArtLine_650.pkl')
m = learn.model
from torch.autograd import Variable
size = 256
dummy_input = Variable(torch.randn(1, 3, 256, 256))
torch_out = torch.onnx._export(
m
, dummy_input
, 'artline_256x256_float32.onnx'
, verbose=True
, opset_version=12
)
/usr/local/lib/python3.6/dist-packages/fastai/vision/models/unet.py:31: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We can't record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
if ssh != up_out.shape[-2:]:
---------------------------------------------------------------------------
RuntimeError Traceback (most recent call last)
<ipython-input-21-4c0daefc8471> in <module>()
7 , 'artline_256x256_float32.onnx'
8 , verbose=True
----> 9 , opset_version=12
10 )
7 frames
/usr/local/lib/python3.6/dist-packages/torch/onnx/symbolic_registry.py in get_registered_op(opname, domain, version)
107 else:
108 msg += "Please open a bug to request ONNX export support for the missing operator."
--> 109 raise RuntimeError(msg)
110 return _registry[(domain, version)][opname]
RuntimeError: Exporting the operator mv to ONNX opset version 12 is not supported. Please open a bug to request ONNX export support for the missing operator.
Oh..., that's very sad, but thank you, you are awsome.
Hello, can you convert please https://github.com/vijishmadhavan/ArtLine to tflite?