Datasetpublic

ENC-PSL/evahan-ultraglyph

Chinese OCR Line Dataset This repository contains the synthetic dataset used by the ENCHANTeam team for the EvaHan 2026 competition on OCR/HTR of Chinese documents. Final ranking of the team: 3rd. UltraGlyph data consist in a automatic mix of real data provided by the organizers, for the closed modality of the competition (no external dataset authorized). Resource Link Preprint HAL Code GitHub Competition EvaHan @ LREC 2026 Dataset description… See the full description on the dataset page: https://huggingface.co/datasets/ENC-PSL/evahan-ultraglyph.

sourceHugging Facecc-by-4.0updated 4mo agoView on Hugging Face

0likes26downloads

Dataset Card

Chinese OCR Line Dataset

This repository contains the synthetic dataset used by the ENCHANTeam team for the EvaHan 2026 competition on OCR/HTR of Chinese documents.

Final ranking of the team: 3rd.

UltraGlyph data consist in a automatic mix of real data provided by the organizers, for the closed modality of the competition (no external dataset authorized).

Resource	Link
Preprint	HAL
Code	GitHub
Competition	EvaHan @ LREC 2026

Dataset description

Pairs of line images and Chinese transcriptions drawn from two sources:

Source label	Origin	Train	Val
`fonts`	Synthetic printed lines using chinese fonts	72 323	12 676
`ultraglyph`	Synthetic lines using real data and ultraglyph generator (mix handwritten and printed)	57 063	10 201
Total		129 386	22 877

Transcription lengths range from 2 to 30 characters (median 16).

Usage

python

from datasets import load_dataset

ds = load_dataset("<org>/<dataset-name>")
example = ds["train"][0]
example["image"].show()
print(example["transcription"], example["source"])

Citation

If you use this dataset, please cite this work:

bibtex

@unpublished{atzori:hal-05622586,
  TITLE = {{Building Character(s): Synthetic Data and In-Context Learning Strategies for Few-Shot Ancient Chinese Recognition}},
  AUTHOR = {Atzori, Denise and Bizais-Lillig, Marie and Garnier, Mathias and L{\'e}toff{\'e}, Maxime and Planque, Charles and Yin, Tianjie and Vidal-Gor{\`e}ne, Chahan},
  URL = {https://enc.hal.science/hal-05622586},
  NOTE = {working paper or preprint},
  YEAR = {2026},
  MONTH = May,
  KEYWORDS = {Historical Chinese OCR ; Synthetic Data ; VLM ; Few-Shot Learning ; In-Context Learning},
  PDF = {https://enc.hal.science/hal-05622586v1/file/paper_short.pdf},
  HAL_ID = {hal-05622586},
  HAL_VERSION = {v1},
}

Acknowledgement

This study was conducted as part of the DH master’s program at École nationale des chartes–PSL and with the support of the PSL Research University’s Major Research Program CultureLab, implemented by the ANR (reference ANR-10-IDEX-0001).