andynoodles/Taiwan-UrbanPlan

台灣都市計畫書 OCR — block-level (zh-Hant)

Block-level OCR pairs synthesised from scanned Taiwanese urban-plan books (都市計畫書) published by county governments. Each row is one cropped layout region with its block type and the vision-LLM transcription — including raw OTSL for tables.

• 119,735 rows from 15,954 pages / 318 plan books (計畫書 + some 計畫圖)
• 36 urban-planning districts (都計區) in 屏東縣 (Pingtung) and 嘉義市 (Chiayi City), plan dates spanning 民國40年代–110年代 (1950s–2020s)
• Mostly photocopier-scanned paper reports (no text layer), rich in statistics tables, land-parcel (地號) lists, zoning tables, and line-drawn maps with text labels
• Source: 縣市都市計畫書圖服務網 public download systems (urbanplanning.pthg.gov.tw, landuse.chiayi.gov.tw)

How it was made

Scanned page → layout detection (PP-DocLayoutV3) → per-region crops → OCR by a PaddleOCR-VL vision-LLM (the teacher) → revision by a Qwen3.6 vision-LLM (the reviser) → Traditional-Chinese normalization (OpenCC s2t). This is a distillation dataset: there is no human ground truth — labels are model output and may contain errors.

Two text fields per block:

• ocr — the teacher's raw output, captured before any rendering post-processing: tables are OTSL (not HTML), everything else plain text.
• ocr_revised — Qwen3.6 correction of the raw OCR against the crop image (populated for 98.6% of rows). The reviser runs with a corpus-specific system prompt (urban-planning vocabulary, ROC dates, land-parcel numbers, photocopy-digit care) plus each book's catalogue metadata (計畫案名, 隸屬都計區, 計畫性質, 民國年 …) from the source system as reference-only context. For tables, a structure guard discards any revision that changes the OTSL grid signature (those rows have ocr_revised = null → use ocr).

Both text fields are normalized to Traditional Chinese with OpenCC (s2t).

Schema

column	type	notes
image	image	the crop fed to the VLM
type	string	layout label (see distribution)
ocr	string	raw teacher output; OTSL for tables
ocr_revised	string	Qwen3.6-corrected text; null → fall back to ocr
bbox	int[4]	[x1,y1,x2,y2] in page pixels
polygon_points	int[][]	null in rectangular layout mode
group_id	int	merged-block grouping
page_width, page_height	int	for normalising bbox
doc_id	string	e.g. urban_plan_pthg_0047_PF000002000087 (site + detail id + file id)
page_index, block_index	int	(doc_id, page_index, block_index) is the key
source	string	urban_plan

Type distribution

text 54.8% · paragraph_title 10.4% · number 10.2% · table 5.4% · image 4.9% · figure_title 4.0% · vision_footnote 3.6% · header 1.9% · footer 1.4% · doc_title 1.3% · seal 1.2% · plus content, chart, footnote, … — a notably table/figure-rich corpus.

Content character

發布實施計畫書 for 都市計畫 main plans and 個案變更: zoning statistics (面積/公頃 tables), 地號 land-parcel lists, population projections, public- facility inventories, 公文 letters with official seals, and line-drawn plan maps with text labels. Older books are typewriter-era photocopies; recent ones are word-processed but still scanned for publication.

Usage

from datasets import load_dataset
ds = load_dataset("andynoodles/TaiwanUrbanPlan", split="train")
tables = ds.filter(lambda r: r["type"] == "table")   # OTSL tables

Quality & limitations

• Model-generated labels (distillation) — not human-verified, grounded in the crop image only (photocopy scans have no text layer).
• Photocopy artifacts (skew, toner noise, fold-out pages) degrade small digits; prefer ocr_revised, and treat number/vision_footnote rows with care.
• table rows: both fields are OTSL; ocr_revised=null (structure-guard discard) is concentrated here — fall back to ocr.
• A handful of giant fold-out map pages exceeded the OCR server's pixel limit; affected blocks carry empty/raw-only text.
• image-type rows are figures/maps; OCR fields carry description-style output, not transcription.

Acknowledgements

• PP-DocLayoutV3 — document layout detection.
• PaddleOCR-VL-1.6 — vision-language OCR (the ocr field).
• Qwen3.6-35B-A3B-FP8 — OCR revision (ocr_revised).
• OpenCC — Simplified→Traditional conversion.

License & provenance

Source books were downloaded from the public 都市計畫書圖服務網 systems of 屏東縣政府 and 嘉義市政府, both released under 政府資料開放授權條款第1版 (Open Government Data License v1.0, Taiwan) — free reuse including commercial, sublicensable, irrevocable, attribution required: 資料來源： 屏東縣政府、嘉義市政府. Plan books are government acts (公文) additionally excluded from copyright by Taiwan Copyright Act §9. Labels are generated by PaddleOCR-VL and Qwen3.6; verify fitness for your use case.

Citation (required)

This dataset is part of the Taiwan Document OCR (zh-Hant) collection by Andy Chen (andynoodles). If you use this dataset — in research, training data, products, or derivatives — you must cite it:

@misc{chen2026_urban_plan,
  title        = {Taiwan Urban Plan Books OCR (block-level, zh-Hant)},
  author       = {Chen, Andy (andynoodles)},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/andynoodles/TaiwanUrbanPlan}},
  note         = {Part of the Taiwan Document OCR (zh-Hant) collection}
}

Citing this dataset is required in addition to the source-attribution noted under License & provenance above. Please retain both this citation and the original source attribution in any redistribution or derived work.