How to Use Prebuilt Deep Learning Algorithms in SageMaker?

Question

Is it possible to use AWS prebuilt algorithms, e.g., Image Classification, locally with SageMaker? I tried to pull the image-classification image URI using aws ecr get-login-password but I get the following error message:

Error response from daemon: pull access denied for 813361260812.dkr.ecr.eu-central-1.amazonaws.com/image-classification, repository does not exist or may require 'docker login': denied: User: xxxxxxxxxxxxxxxxxxxxxxxx is not authorized to perform: ecr:BatchGetImage on resource: arn:aws:ecr:eu-central-1:813361260812:repository/image-classification because no resource-based policy allows the ecr:BatchGetImage action

I gave my user full access so the policies include BatchGetImage. Can someone explain why this error occurs?

Thanks!

user_9805457 · Answer

Try using the local mode instead. Otherwise have a look at the many examples provided here (including custom containers): https://github.com/aws-samples/amazon-sagemaker-local-mode

user_7526450 · Answer

In my particular case, I get the same error when the URl was bad. On ECR, go to your repo, on Images, Image URI, click on Copy URI and try again, nothing related to resources on IAM. For this same reason, check if sagemaker is not changing the URI.

If you are doing a cross account access recall to add proper policy on repo.

How to Use Prebuilt Deep Learning Algorithms in SageMaker?

2 Answers