Datasets:

HuggingFaceFW
/

ocr-annotations

dataset_info:
  features:
    - name: filename
      dtype: string
    - name: pdf
      dtype:
        pdf:
          decode: false
    - name: annotation
      dtype: string
    - name: class
      dtype: string
    - name: truncation_type
      dtype: string
    - name: pdf_size_bytes
      dtype: int64
  splits:
    - name: train
      num_bytes: 1503710871
      num_examples: 1620
  download_size: 1410319327
  dataset_size: 1503710871
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

PDF OCR Classification Dataset

This dataset contains PDF documents with annotations for OCR classification tasks.

Dataset Description

Total samples: 1620
Classes: OCR (requires OCR processing), NOCR (no OCR needed)

Dataset Structure

Each row contains:

filename: Original PDF filename
pdf: PDF file as binary data (using Pdf feature type)
annotation: Original annotation label
class: Binary classification label (OCR/NOCR)
truncation_type: Whether the PDF is truncated or non-truncated
pdf_size_bytes: Size of the PDF file in bytes

Class Distribution

class NOCR 1393 OCR 227

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("HuggingFaceFW/ocr-annotations")

# Access train split
train_data = dataset['train']

# Access a sample
sample = train_data[0]
pdf_bytes = sample['pdf']  # This will be bytes
label = sample['class']

License

Please check the original data source for licensing information.