chain VLM + image classification

[geoHeil]

How can I chan the VLM pipeline (smoldocling, granite) with the picture classification capability? https://docling-project.github.io/docling/usage/enrichments/#formula-understanding

pipeline_options.do_picture_classification = True this flag is currently not supported for the VlmPipelineOptions

[dolfim-ibm]

Not yet, but it might not be needed. SmolDocling has the picture classification built-in in the predictions.

[geoHeil]

I have seen the separate smoldocling classifier. But so far did not find a way to enable that.

the default VlmPipelineOptions do not compute/store these classifications. Am I overlooking something is this still a feature request? Especially if these are anyways part of smoldocling - it would be very neat if they would be stored as well

[geoHeil]

And otherwise if it is a 2- step process some clarifying documentation - such as for https://github.com/DS4SD/MolGrapher?tab=readme-ov-file#docling-integration for the enrichment step that combines both tasks would be highly appreciated