Error executing Auto ML Job for image task

Onyango, David 31 Reputation points

I am getting this error when running a job an AutoML job for images.

The data type of image URL column 'image_url' is NULL, but it should be STREAM. Marking the experiment as failed because initial child jobs have failed due to user error

I have used 2 MLTable formats image tasks as follows unsuccessfully.

Format 1

paths:

  • file: ./test.jsonl

transformations:

  • read_json_lines: encoding: utf8
  • convert_column_types:
    • columns: image_url column_type: stream_info
    • columns: labels column_type: string

Format 2

paths:

  • file: ./val.jsonl

transformations:

  • read_json_lines: encoding: utf8
  • convert_column_types:
    • columns: image_url column_type: stream
    • columns: labels column_type: string

The corresponding .jsonl files have each line in the form below.

{"image_url": "img_0.jpg", "labels": "Unknown"}

NOTE: For background information, I am migrating from Custom Vision to Auto ML. I have successfully moved the labelled images from custom vision to my blob storage container.

  1. Anshika Varshney 13,405 Reputation points • Microsoft External Staff • Moderator

    Hello Onyango, David,

    Thanks for the clarification in the private chat. Even though the job is submitted from the Azure ML Studio UI, The UI itself is only a control plane. Automated ML jobs do not run in the browser they execute on a backend compute target and managed environment.

    When an AutoML Image job starts, Azure ML automatically provisions a compute cluster, creates a Docker-based environment, and installs required dependencies. If any of these steps fail, the job will error out even though the UI appears healthy.

    The best place to start is the failed job’s Logs section in Azure ML Studio. Please check the files under azureml-logs/, especially 70_driver_log.txt and stderr_log.txt, as they contain the actual root cause (environment creation, dependency install, CUDA/GPU, or Docker pull issues).

    Also verify that the selected compute cluster is running, supports GPU, and is available in your workspace region, as AutoML Image tasks have region and VM-size limitations.

    In short: The issue is not with the GUI itself, but with the execution environment behind the scenes. The logs will clearly indicate which component is failing.
    I hope this has been helpful!
    Thankyou!

Sign in to comment

1 answer

  1. Anshika Varshney 13,405 Reputation points • Microsoft External Staff • Moderator

    Hi Onyango, David,

    Welcome to Microsoft Q&A and Thank you for reaching out.

    This error is commonly seen with AutoML image jobs and is usually related to the execution environment rather than the image task itself.

    A few things the community typically checks first:

    Environment / dependency issues: AutoML image jobs auto-generate a Conda environment, and failures often occur due to package conflicts or unsupported Python versions. Sticking to Python 3.8–3.10 usually helps.

    Image build failures: Sometimes the job fails while building the Docker image (dependency resolution, base image pull issues, or deprecated images).

    Dataset or compute setup: Verify that the image dataset format matches the task (classification/detection/segmentation) and that the compute target has sufficient resources (especially GPU).

    Next steps

    1. Check the detailed job logs in the Azure ML portal (or download them via CLI) the root cause is usually clearly logged there.
    2. If the failure is environment-related, try using a custom environment instead of the fully auto-generated one.
    3. Confirm that your Azure ML SDK, workspace, and compute are on supported versions.

    Please let me know if there are any remaining questions or additional details, I can help with, I’ll be glad to provide further clarification or guidance.

    Thnakyou!

    1. Anshika Varshney 13,405 Reputation points • Microsoft External Staff • Moderator

      Hi Onyango, David,

      I’ve converted my earlier comment into an answer, could you please take a moment to mark it as Accepted? This helps others in the community with the same question find the solution more easily.

      👁 User's image

    Sign in to comment
Sign in to answer

Your answer