datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
japanese-str-dataset-v1
STR Dataset
Japanese STR (Scene Text Recognition) dataset in WebDataset format.
This dataset is composed of:
Images of Japanese named entities (full names and their affiliations)
Images of sentences retrieved from Aozora Bunko (青空文庫)
and their corresponding ground truth texts.
All images are synthesized using TRDG.
Dataset Structure
Split
Samples
Shards
train
10,000,000
1000
valid
50,000
5
test
50,000
5
Total
10,100,000
1010
Usage
import… See the full description on the dataset page: https://huggingface.co/datasets/nagohachi/japanese-str-dataset-v1.jawildtext_cropped
jawildtext_cropped
Per-polygon scene-text-recognition crops derived from
llm-jp/jawildtext.
Each row of the source dataset carries a polygons column with quadrilateral
text regions and their transcriptions. For every polygon we perspective-warp
the source image onto the rectified bounding rectangle, yielding a tight,
horizontally-aligned crop suitable for training/evaluating Japanese scene-text
recognition models.
Stats
Samples: 108 403
Shards: 22… See the full description on the dataset page: https://huggingface.co/datasets/nagohachi/jawildtext_cropped.NDL_pdm-ocr-part2_cropped
NDL Public Domain OCR Dataset — Line-Cropped (WebDataset)
This dataset is a derivative of the Public Domain OCR Training Dataset (FY2021) published by the National Diet Library of Japan (国立国会図書館).
Each sample is a line-level crop of a historical document page, paired with its OCR text transcription.
What was changed from the original
This dataset was not created by the National Diet Library. The following transformations were applied:
Each text line (LINE element)… See the full description on the dataset page: https://huggingface.co/datasets/nagohachi/NDL_pdm-ocr-part2_cropped.