aws-sagemaker

Viam provides an aws-sagemaker model of vision service with which you can use ML models you deploy on cloud endpoints using AWS Sagemaker.

Configure this vision service as a modular resource on your robot to access and perform inference with AWS-deployed ML models.

Build and Run

The first step is to deploy your model to the AWS Sagemaker endpoint. You can deploy programmatically but we recommend doing it through the AWS console. Instructions for setting up an endpoint (using a pre-trained Sagemaker Jumpstart model) are as follows:

• Log in to the AWS console as a root user or a user with administrator access
• At the Console Home, in the search bar, find and click Amazon Sagemaker
• Follow the instructions under “Getting started” to create a domain
• Launch your domain with Sagemaker Studio, which will open up JupyterLab in a new tab
• Go to the new tab. On the left bar, click Sagemaker Jumpstart > Models, notebooks, solutions
• (To sort the pre-trained models by task, click the ML tasks tab)
• Select a model under “Image Classification” or “Object Detection” (for classification or detection, respectively)
• Before deploying, under “Deployment Configuration”, choose the SageMaker hosting instance appropriate for your model (usually ml.m4.xlarge or ml.m5.xlarge)
• Finally, click Deploy! After a few moments your endpoint will be ready for inference.

For more detailed instructions, click here.

Note

Before configuring your vision service, you must create a robot.

Configure your aws-sagemaker vision service

Navigate to the Config tab of your robot’s page in the Viam app. Click on the Services subtab and click Create service. Select the vision type, then select the aws-sagemaker model. Enter a name for your service and click Create.

On the new service panel, copy and paste the following attribute template into your service’s Attributes box.

{
  "access_json": "/Users/myname/Documents/accessfile.json",
  "endpoint_name": "",
  "aws_region": "",
  "source_cams": [
    "myCam1",
    "myCam2"
  ]
}

Edit and fill in the attributes as applicable and create an access JSON file.

Note

For more information, see Configure a Robot.

Attributes

The following attributes are available for viam:vision:aws-sagemaker vision services:

Name	Type	Inclusion	Description
endpoint_name	string	Required	The name of the endpoint as given by AWS.
aws_region	string	Required	The name of the region in AWS under which the model can be found.
access_json	string	Required	The on-robot location of a JSON file that contains your AWS access credentials (AWS Access Key ID and Secret Access Key). Follow these instructions to retrieve your credentials, and reference this example JSON file containing credentials.
source_cams	array	Required	The name of each camera you have configured on your robot that you want to use as input for the vision service.

Example Access JSON

{
  "access_key": "UE9S0AG9KS4F3",
  "secret_access_key": "L23LKkl0d5<M0R3S4F3"
}

Example Configuration

{
  "modules": [
    {
      "type": "registry",
      "name": "viam_aws-sagemaker",
      "module_id": "viam:aws-sagemaker",
      "version": "0.0.1"
    }
  ],
  "services": [
    {
      "name": "myVisionModule",
      "type": "vision",
      "namespace": "rdk",
      "model": "viam:vision:aws-sagemaker",
      "attributes": {
        "access_json": "/Users/myname/Documents/accessfile.json",
        "endpoint_name": "jumpstart-dft-tf-ic-efficientnet-b1-classification-1",
        "aws_region": "us-east-2",
        "source_cams": [
          "myCam1",
          "myCam2"
        ]
      }
    }
  ]
}

Next steps: use your ML-enabled vision service

Now, use the viam:vision:aws-sagemaker modular service to perform inference with the machine learning model deployed through AWS Sagemaker on your robot.

Configure a transform camera to see classifications or detections appear in your robot's field of vision.

You can also use the following methods of the vision service API to programmatically get detections and classifications with your modular vision service and camera:

• GetDetections()
• GetDetectionsFromCamera()
• GetClassifications()
• GetClassificationsFromCamera()