Soxavin/ardb-gemma-sft-v2

Note: A newer, larger version of this dataset lineage is available at Soxavin/ardb-sft-v5 — same task design (one page image in, boxes + text out in a single forward pass), expanded from a single-year corpus (28 documents) to a multi-year one (47 documents, 2022–2026), with an era-stratified train/validation/test split. This version is kept as-is for reproducibility of results computed against these exact 66/9/9 rows; use v5 for new work.

ARDB Unified Page Understanding (bbox + text, one forward pass)

Vision fine-tuning data for Gemma 4 E2B (via Unsloth), built from ARDB (Agricultural and Rural Development Bank of Cambodia) daily market-price bulletins. This dataset unifies layout detection (bbox only) and transcription (text only, on pre-cropped regions): one full page image in, one JSON list out, where each region carries its box AND its transcribed text together, in a single forward pass.

Stats

Split	Rows (pages)	Docs
train	66	22
validation	9	3
test	9	3

28 documents total (2 evaluation bulletins reserved for other tracks, excluded here).

Schema

Field	Type	Description
image	image	Full rendered page
instruction	string	Fixed task instruction
text	string	JSON list of {"box_2d": [y1,x1,y2,x2], "label": category, "text": content}
doc_id	string	Document identifier
source	string	Source PDF filename
page	int	Page index (0-based)
date	string	Bulletin date (page 0 only, else empty)

box_2d is normalized to a 0-1000 grid (Gemma's native detection convention), and the list is sorted top-to-bottom by box_2d[0]. Categories: Table, Text, Section-Header, Page-Furniture, Picture. text is "" for regions with no content to transcribe (Picture).

Construction

Two independent, already-verified pipelines, zipped by (page, label):

• Boxes come straight from ardb-layout-coco-v2's human-corrected COCO annotations (no new annotation work).
• Text comes from template substitution: every ARDB bulletin shares one fixed layout, so Table and Section-Header (title) text is built by verifying each layout once and substituting each document's own varying content (table prices/dates, the title's day and month), rather than running OCR independently per document. Page-Furniture (letterhead) and Text (source-note footer) are fully static across every document and were verified directly against rendered crops.

A page is skipped entirely (never emitted with one field silently wrong) if its table can't be aligned to the template, or a Section-Header/Page-Furniture region can't be resolved to text.

License

cc-by-nc-4.0. Images are reproductions of publicly published ARDB bulletins (research use only; not affiliated with or endorsed by ARDB).