Hey,
I'm trying to use Ground Truth to do image classification but with a different set of label options for each image. I have the custom labeling task template and pre-/post-labeling Lambda functions set up and I figured I could pass in the labels through the manifest file.
My issue is that the Ground Truth job ignores the attributes in the manifest file that are not "source-ref" (or "source"). This causes the pre-processing Lambda function to fail because the request it is passed only contains the "source-ref" attribute, but the Lambda function also references a different attribute. Are augmented manifest files supported for Ground Truth and if they are, how can I make use of the extra attributes?
References:
Ground Truth Input Data: https://docs.aws.amazon.com/sagemaker/latest/dg/sms-data-input.html
SageMaker Augmented Manifest Files: https://docs.aws.amazon.com/sagemaker/latest/dg/augmented-manifest.html
Example:
A normal Ground Truth manifest file:
{"source-ref":"s3://some_bucket/images/img1.png"}
{"source-ref":"s3://some_bucket/images/img2.png"}
...
What I want to be able to use:
{"source-ref":"s3://some_bucket/images/img1.png","labels":["pen","pencil","stick"]}
{"source-ref":"s3://some_bucket/images/img2.png","labels":["tv","laptop","phone"]}
...
Is there any update on this?