Felix92/ibm-handwriting-campaign-word

IBM Handwriting Campaign Word Dataset

Source dataset: docling-project/ibm-handwriting-campaign

This directory contains a Hugging Face dataset export generated from the project source handwriting data.

Overview

• Splits: train, validation, test
• Total samples: 841 (train: 671, validation: 83, test: 87)
• Format: Parquet files with images stored as bytes

Each record corresponds to a single word extracted from scanned forms and includes the word image, annotation JSON, transcript text, and a resolved writer_id.

Features

• id: unique sample identifier (built from writer folder + file stem)
• image: word image bytes
• original_path: relative path inside the source dataset
• annotation: parsed JSON annotation for the word
• text: word transcript
• writer_id: numeric writer identifier

Source layout

Each sample typically has an image, an annotation JSON, and a transcript text file. Split lists in this project may reference base form names and variants (for example _top, _bottom).

Output

The exported files are located under data/:

• data/train/train-0.parquet
• data/validation/validation-0.parquet
• data/test/test-0.parquet
• data/metadata.json (dataset-level metadata)

Usage

If the dataset is hosted on the Hugging Face Hub, load a split directly:

from datasets import load_dataset

ds = load_dataset("docling-project/ibm-handwriting-campaign-word", split="train")
print(ds[0]["text"])

Notes

• writer_id is provided as a numeric identifier for each sample.
• The text column contains the word transcript as plain text for convenient access and quick filtering/searching.
• Images are embedded as bytes in the parquet files so the export can be uploaded to the Hugging Face Hub without separate image files.