How to use Ruby to send image to a deployed Sagemaker endpoint running a TensorFlow/Keras CNN? - ruby

Question

I have trained a CNN using Tensorflow/Keras and successfully deployed it to Sagemaker using the saved_model format. It answers pings and the dashboard shows it is running.
I now need to be able to send it images and get back inferences. I have already successfully deployed an ANN to Sagemaker and gotten predictions back, so most of the "plumbing" is already working.
The Ruby performing the request is as follows:
def predict
sagemaker = Aws::SageMakerRuntime::Client.new(
access_key_id: Settings.sagemaker_key_id,
secret_access_key: Settings.sagemaker_secret,
region: Settings.sagemaker_aws_region
)
response = sagemaker.invoke_endpoint(endpoint_name: Settings.sagemaker_endpoint_name,
content_type: 'application/x-image',
body: File.open('developer/ai/caox_test_128.jpg', 'rb'))
return response[:body].string
end
(For now, I simply hardcoded a known file for testing.)
When I fire this, I get back this error: Aws::SageMakerRuntime::Errors::ModelError: Received client error (400) from model with message "{ "error": "JSON Parse error: Invalid value. at offset: 0" }"
It's almost as if the model is expecting more in the body than just the image, but I can't tell what. AWS's documentation has an example for Python using boto:
import boto3
import json
endpoint = '<insert name of your endpoint here>'
runtime = boto3.Session().client('sagemaker-runtime')
# Read image into memory
with open(image, 'rb') as f:
payload = f.read()
# Send image via InvokeEndpoint API
response = runtime.invoke_endpoint(EndpointName=endpoint, ContentType='application/x-image', Body=payload)
# Unpack response
result = json.loads(response['Body'].read().decode())
As far as I can tell, they are simply opening a file and sending it directly to sagemaker with no additional pre-processing. And, insofar as I can tell, I'm doing exactly what they are doing in Ruby, just using 'aws-sdk'.
I've looked through Amazon's documentation, and for examples on Google, but there is scant mention of doing anything special before sending the file, so I'm scratching my head.
What else do I need to consider when sending a file to a Sagemaker endpoint running a TensorFlow/Keras CNN to get it to respond with a prediction?