Document Layout Test A
About
Polish Language Document Layout Detection — Test A
PolEval 2026 task: find the layout regions on scanned Polish document pages and label each with one of 11 categories: Caption, Footnote, Formula, List-item, Page-footer, Page-header, Picture, Section-header, Table, Text, Title. Page images and the task description: https://github.com/poleval/2026-document-layout
The inputs list one page per line:
id: the page identifier a submission must use.image_id: the page's COCO image id.file_name: the page image in the organisers' repository, relative topoleval_doclaynet_coco_test_a/.width,height: the coordinate frame boxes are given in.
Submissions are one JSONL line per page, {"id": "<page id>", "output": [{"category", "bbox": [x, y, w, h], "score"}]}. They are scored by COCO mAP@[.50:.95], the mean of the 11 per-category APs.
The data is converted from the organisers' COCO files without changing any value. 37 pages are listed as 1×1, with fractional box coordinates, exactly as the organisers published them.
Campaigns
Inputs
hf download ipipan/llms4eu-benchmarks-inputs \ document_layout_test_a/inputs.jsonl \ document_layout_test_a/manifest.json \ --repo-type dataset \ --revision document_layout_test_a-1.0.0
Definition
An example item
{
"id": "20b45f96-3093-5c43-9dc4-08c938b29f96",
"image_id": 19,
"file_name": "images/Etyka 2023_00.jpg",
"width": 1474,
"height": 2048
}
Example output
A submission line for an item like this: one JSON line with its id and the model's raw output. The answer key is private.
{"id": "20b45f96-3093-5c43-9dc4-08c938b29f96", "output": [{"category": "Text", "bbox": [72.0, 96.5, 451.0, 38.0], "score": 0.93}]}