onnx_cpu

onnx-cpu modular resource

This module allows you to deploy ONNX models which you can then use with the Viam mlmodel vision service on Android.

Requirements

Before configuring your ML model, you must create a machine.

Build and run

To use this module, follow these instructions to add a module from the Viam Registry and select the viam-labs:mlmodel:onnx-cpu model from the onnx-cpu module.

Configure your onnx-cpu ML model

Navigate to the Config tab of your machine's page in the Viam app. Click on the Services subtab and click Create service. Select the mlmodel type, then select the onnx-cpu:timesyncsensor model. Click Add module, then enter a name for your ML model and click Create.

On the new component panel, copy and paste the following attribute template into your ML model’s Attributes box:

{
  "model_path": "/path/to/onnx_file/detector_googlenet.onnx",
  "label_path": "/path/to/labels.txt"
}

[!NOTE] For more information, see Configure a Machine.

Attributes

The following attributes are available for the viam-labs:mlmodel:onnx-cpu ML model service:

Example configuration

    {
      "name": "onnx",
      "type": "mlmodel",
      "model": "viam-labs:mlmodel:onnx-cpu",
      "attributes": {
        "model_path": "/path/to/onnx_file/detector_googlenet.onnx"
        "label_path": "/path/to/labels.txt"
      }
    }

Configure a vision service

The module allows you to deploy an ONNX ML model with the ML model service it provides. Viam's vision service allows you to make use of the deployed detection or classification model.

Navigate to the Config tab of your machine's page in the Viam app. Click on the Services subtab and click Create service. Select the vision type, then select the mlmodel model. Enter a name for your ML model and click Create.

On the new service panel, select the onnx-cpu model you configured.

You will most likely need to rename the input and outputs tensors coming from the ONNX file to use the tensor names that the vision service requires. To rename the tensors, go to your Raw JSON configuration and add the remap_output_names and remap_input_names fields to the attributes of your vision service config. For more information see the documentation for remapping tensor names.

Here is an example:

   {
      "type": "vision",
      "model": "mlmodel",
      "attributes": {
        "mlmodel_name": "onnx",
        "remap_output_names": {
          "detection_classes": "category",
          "detection_boxes": "location",
          "detection_scores": "score"
        },
        "remap_input_names": {
          "input_tensor": "image"
        }
      },
      "name": "onnx-vision"
    }

Save your config, before testing your vision service.

Next steps

You have now configured a vision service to use an ONNX ML model. Follow these steps to test your mlmodel vision service.

Local Development

This module is written in Go. If you need to package it up into a binary to create a module, run the following commands:

go build -o module cmd/module/main.go
chmod a+x module
tar -czf module.tar.gz module third_party/

third_party contains all of the .so files for the ONNX runntime. You can package only the one you need.

For Android, the Makefile will create the tar file.

You can then add the module to app.viam.com locally. the model triplet is viam-labs:mlmodel:onnx-cpu.

License

Copyright 2021-2023 Viam Inc.
Apache 2.0