The problem you have encountered:

When you import new data into an existing managed vertex dataset a new annotation set is created every time. Even when setting "annotationResourceLabels":{"aiplatform.googleapis.com/annotation_set_name":"XXX"}} in the .jsonl file. This property is ignored and instead, a new annotation set is created.

This happens using the SDK (1.10.0) and using the REST API. When importing data in the UI data is correctly added to the "active" annotation set.

This results in the fact that it is impossible to programmatically update your dataset.

What you expected to happen:

When explicitly setting the target annotation set one would expect the data to get added to that set.

Steps to reproduce:

• create dataset.jsonl file and upload it to gcs
• execute using SDK (1.10.0):

dataset.import_data(
    gcs_source=[annotation_file_uri],
    import_schema_uri=aiplatform.schema.dataset.ioformat.image.bounding_box
)

Other information (workarounds you have tried, documentation consulted, etc):

Since I got the expected behavior through the UI I figured I'd give REST requests a try. But this also has the unwanted behavior of creating a new annotation set and ignoring the explicitly set one.