ONNX not supported Mac OS X

[isaacmg]

I'm trying to test out the ONNX preset on Mac and I'm getting the following error.

Exception in thread "main" java.lang.UnsatisfiedLinkError: Platform "macosx-x86_64" not supported by class org.bytedeco.javacpp.onnx
	at org.bytedeco.javacpp.Loader.load(Loader.java:928)
	at org.bytedeco.javacpp.Loader.load(Loader.java:891)
	at org.bytedeco.javacpp.onnx.<clinit>(onnx.java:10)
	at java.lang.Class.forName0(Native Method)
	at java.lang.Class.forName(Class.java:348)
	at org.bytedeco.javacpp.Loader.load(Loader.java:950)
	at org.bytedeco.javacpp.Loader.load(Loader.java:891)
	at org.bytedeco.javacpp.onnx$ISchemaRegistry.<clinit>(onnx.java:1324)
	at LoadModelONNX.<init>(LoadModelONNX.java:6)
	at runModel.main(runModel.java:6)

This is what my POM file looks like.

<?xml version="1.0" encoding="UTF-8"?>
<project xmlns="http://maven.apache.org/POM/4.0.0"
         xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
         xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
    <modelVersion>4.0.0</modelVersion>

    <groupId>org.model_loader</groupId>
    <artifactId>caffe_model</artifactId>
    <version>1.0-SNAPSHOT</version>
    <dependencies>
        <!-- https://mvnrepository.com/artifact/org.bytedeco.javacpp-presets/onnx-platform -->
        <dependency>
            <groupId>org.bytedeco.javacpp-presets</groupId>
            <artifactId>onnx-platform</artifactId>
            <version>1.3.0-1.4.3</version>
        </dependency>

    </dependencies>

</project>

Am I missing anything?

[saudet]

No, we probably just need to adjust the build in the cppbuild.sh file:
https://github.com/bytedeco/javacpp-presets/blob/master/onnx/cppbuild.sh
Could you give it a try and send a pull request if it works? That would be great!

[isaacmg]

Sorry, I'm not too familiar with javacpp-presets or the maven build process (I mostly work in Python but need to make sure our model runs in Java). Are you saying to download the sh file and run it manually on my machine or something else? Thanks.

[saudet]

Yes, that's basically it: https://github.com/bytedeco/javacpp-presets#build-instructions And we'll also need to add "macosx" next to "linux" in the presets/onnx.java file, but that should be it.

[isaacmg]

Okay I added "macosx" to presets/onnx.java, but when went to install with maven I ran into fatal error: too many errors emitted, stopping now I think this is a problem with my Mac though as it has happened before when building caffe2 from the source.

[saudet]

I've included support for Mac in this PR: #622
Please give it a try! Thanks

[isaacmg]

Thanks I'm still having problems building on my Mac but I think this is due to problems with cmake on my end. I did get it running on Linux. I was wondering once I load the model what method do I call to actually make the predictions? Usually in Python I would do something like

prepared_backend = onnx_caffe2.backend.prepare(model)
W = {model.graph.input[0].name: model_inputs.numpy()}
c2_out = prepared_backend.run(W)[0]

Now I just have a ModelProto class but don't know what to do with it.

[saudet]

I've merged the PR, so snapshots will be available sometime tomorrow.
Please try those: http://bytedeco.org/builds/ Thanks!

[saudet]

@EmergentOrder Any ideas about how we can do what @isaacmg mentions above?

[EmergentOrder]

@isaacmg This preset does not include a backend. It enables us to do things like load a model, check it, optimize it, perform shape inference, convert it to a different ONNX version and gain access to the operator schemas and IR.

The JavaCPP Preset offers an API that is as close as possible to the ONNX C++ API in Java.

In contrast, my project ONNX-Scala aims to provide a typeful, idiomatic and purely functional Scala API to ONNX.

ONNX-Scala provides a clean separation between API, program definition, and backend implementation.

Currently there is no backend available for ONNX-Scala either, but the plan is to have several backend implementations available. The first cut of the first one should be ready in the near future.
Lantern is the top candidate right now.

I am also working on a JavaCPP Preset for ngraph , which should provide us with an ONNX backend in Java (also a candidate for a backend in ONNX-Scala)

[saudet]

Support for Mac has been included in version 1.4.4! Enjoy