Everyone's talking about VLMs parsing documents end to end.
But before a page can be read, something has to know where the tables, headers, blocks, and formulas actually are.
That's Object Detection.
Get the boxes wrong and everything downstream gets messy.
Bad crops.
Brittle prompts.
Bloated cleanup code.
Worse OCR, reading order, and table extraction.
So we went deep on fine tuning our own OD model 🤓
Turns out, the “just add data and train” story is... not quite it.
Full write up:
unstructured.io/blog/why-fin…