[FEAT] Add the block type Field for form fields
Author: mardukbpCreated Aug 18, 2026Updated Aug 18, 2026
Labelsenhancement
✨ Is your feature request related to a problem?
Given a PDF that contains the following form
the Chunks output contains the blocks
{
"id": "/page/0/Text/14",
"block_type": "Text",
"html": "<p><b>Mietbeginn:</b></p>",
"page": 24,
"polygon": [
[67.32000000000001, 214.63200000000003],
[128.51999999999998, 214.63200000000003],
[128.51999999999998, 229.67999999999995],
[67.32000000000001, 229.67999999999995]
],
"bbox": [
67.32000000000001, 214.63200000000003, 128.51999999999998,
229.67999999999995
],
"section_hierarchy": {
"1": "/page/0/SectionHeader/10"
},
"images": {}
},
{
"id": "/page/0/Text/15",
"block_type": "Text",
"html": "<p>D3</p>",
"page": 24,
"polygon": [
[244.8, 209.88000000000002],
[264.384, 209.88000000000002],
[264.384, 223.344],
[244.8, 223.344]
],
"bbox": [244.8, 209.88000000000002, 264.384, 223.344],
"section_hierarchy": {
"1": "/page/0/SectionHeader/10"
},
"images": {}
}Assigning the block type Text to both blocks is not correct. The first block is a Text, but the second block is a Field. This distinction is important in order to build a semantic model from a PDF form.
Describe the Solution You'd Like
Editable form fields in a PDF should have the block type Field. They are not read-only text.
Alternatives Considered
From the bounding boxes I can infer which block is a Text and which one is a Field, but marker should do the correct classification.
Additional Context
NA
Source: datalab-to/marker