5
$\begingroup$

the model was trained by

net = NetModel["Inception V1 Trained on ImageNet Competition Data"]

Export["model.json",model,"MXNet"]
model.json
 Import["model.json",{"MXNet","Net"}]
During evaluation of Import::invmxlayer: Layer named "1.pool1_3x3_s2_pad$0" couldn't be created, layer type "Pad" is not supported.
$Failed

So sad

So we cann't import many models into Mathematica.

$\endgroup$
  • $\begingroup$ Since the documentation here declare it is a invertible convert.then I don't think something wrong. $\endgroup$ – yode Sep 16 '17 at 8:11
  • $\begingroup$ @yoda bad news. I think Mathematica implements some stat-of-art models, should support this manipulation as new features.. $\endgroup$ – HyperGroups Sep 16 '17 at 8:15
  • $\begingroup$ You didn't spell @yode 's name correctly. Maybe you can try again? $\endgroup$ – QuantumDot Sep 16 '17 at 19:36
  • $\begingroup$ @QuantumDot Copy yours. :) $\endgroup$ – yode Sep 16 '17 at 19:51
4
$\begingroup$

There are two function GeneralUtilities`MLExport and GeneralUtilities`MLImport to do export and import thing.But it cannot produce that .params file.Maybe it can help.

net = NetModel["Inception V1 Trained on ImageNet Competition Data"];
GeneralUtilities`MLExport["test.json", net]

"test.json"

Then you can import it.

netin = GeneralUtilities`MLImport["test.json"]

$\endgroup$
  • $\begingroup$ Nice find. I can't wait for ONNX support. $\endgroup$ – M.R. May 18 '18 at 17:50

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.