I’m wondering if the Document AI Custom Splitter allows me to achieve my goal.
I have a single page file (PDF, or image) with multiple documents on it. E.g. passport, id card, photo 3x4.
My goal is to detect each document and process it separately. So from the example with three documents on the single page, I want to extract each document in a separate file.
I trained a custom splitter but the training process doesn’t include any regions/boundary drawing and only has pages labeling.
Am I right, that Custom Splitter can’t fulfill my requirements?
If so, could you suggest right tool, please?
Thank you in advance.
The current Custom Splitter in Document AI doesn’t support region-level or bounding-box-based splitting within a single page. It’s built around page-level classification, so when your file is a single page with multiple logical documents (like passport + ID + photo), it won’t be able to split them as separate units.
For your use case, what you’re really looking for is region detection and cropping, not just page-level splitting. Unfortunately, that’s outside the scope of Custom Splitter right now.
You’ve got two realistic options:
- Use Document AI with a custom processor built on Layout/DocParser + Vision API - combine text layout understanding with image object detection.
- Or go more hands-on with Vertex AI + AutoML Vision Object Detection, where you train a model to detect the regions (passport, photo), extract their bounding boxes, and then slice them into separate files programmatically.
It’s more effort upfront, but it gives you the control you need for region-based segmentation.