Dataset Viewer
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
image: string
label: string
to
{'image': Image(mode=None, decode=True, id=None)}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 99, in get_rows_or_raise
                  return get_rows(
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 77, in get_rows
                  rows_plus_one = list(itertools.islice(ds, rows_max_number + 1))
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/iterable_dataset.py", line 2285, in __iter__
                  for key, example in ex_iterable:
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/iterable_dataset.py", line 1856, in __iter__
                  for key, pa_table in self._iter_arrow():
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/iterable_dataset.py", line 1888, in _iter_arrow
                  pa_table = cast_table_to_features(pa_table, self.features)
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/table.py", line 2215, in cast_table_to_features
                  raise CastError(
              datasets.table.CastError: Couldn't cast
              image: string
              label: string
              to
              {'image': Image(mode=None, decode=True, id=None)}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

ReceiptSense: Beyond Traditional OCR - A Dataset for Receipt Understanding

Paper Dataset License

🔥 News

  • [2024] ReceiptSense dataset is now publicly available!
  • [2024] Paper accepted and published

📖 Abstract

Multilingual OCR and information extraction from receipts remains challenging, particularly for complex scripts like Arabic. We introduce ReceiptSense, a comprehensive dataset designed for Arabic-English receipt understanding comprising:

  • 20,000 annotated receipts from diverse retail settings
  • 30,000 OCR-annotated images
  • 10,000 item-level annotations
  • 1,265 receipt images with 40 question-answer pairs each for Receipt QA

The dataset captures merchant names, item descriptions, prices, receipt numbers, and dates to support object detection, OCR, information extraction, and question-answering tasks. We establish baseline performance using traditional methods (Tesseract OCR) and advanced neural networks, demonstrating the dataset's effectiveness for processing complex, noisy real-world receipt layouts.

🎯 Key Features

Multilingual Support

  • Arabic-English bilingual receipts
  • Real-world mixed-language content
  • Complex script handling for Arabic text

📊 Comprehensive Annotations

  • Object Detection: Bounding boxes for key receipt elements
  • OCR: Character and word-level text recognition
  • Information Extraction: Structured data extraction
  • Receipt QA: Question-answering capabilities

🏪 Diverse Retail Environments

  • Supermarkets and grocery stores
  • Restaurants and cafes
  • Clothing and retail shops
  • Various geographical regions

🔧 Real-world Challenges

  • Noisy and degraded image quality
  • Complex receipt layouts
  • Mixed fonts and orientations
  • Authentic retail scenarios

📈 Dataset Statistics

Component Training Validation Test Total
Key Information Detection 12,600 3,700 3,700 20,000
OCR Dataset 21,000 4,500 4,500 30,000
Item Information Extraction 7,000 1,500 1,500 10,000
Receipt QA - - 1,265 1,265

Language Distribution

  • Arabic: 53.6%
  • English: 26.2%
  • Mixed Language: 20.3%

Receipt QA Coverage

  • Merchant/Payment/Date Metadata: 30%
  • Item-level Information: 50%
  • Tax/Total/Payment Details: 20%

🖼️ Sample Images

Sample 1 Sample 2 Sample 3 Sample 4 Sample 5

Examples of annotated receipt images showcasing the variety of formats, languages, and complex text layouts

🎯 Supported Tasks

1. 🎯 Key Information Detection

Extract essential receipt information including:

  • Merchant names
  • Transaction dates
  • Receipt numbers
  • Item lists and descriptions
  • Total amounts

2. 🔍 OCR (Optical Character Recognition)

Box-level text annotations for:

  • Multilingual text recognition
  • Complex layout understanding
  • Noisy image processing

3. 📝 Information Extraction

Detailed item-level analysis:

  • Item names and descriptions
  • Prices and quantities
  • Categories and classifications
  • Brands and packaging information

4. ❓ Receipt Question Answering

Comprehensive QA capabilities covering:

  • Receipt metadata queries
  • Item-specific questions
  • Transaction summary questions
  • Payment and tax information

📥 Download Links

🎯 Key Information Detection

🔍 OCR Dataset

📝 Item Information Extraction

❓ Receipt Question Answering

⚠️ Note: All receipt datasets have been updated to include PII-redacted versions for privacy protection.

🏆 Baseline Results

Object Detection Performance

Model Backbone Precision Recall mAP50 mAP50-95
YOLOv7 - 76.0% 85.6% 79.2% 43.7%
YOLOv8 - 74.6% 81.0% 76.1% 45.3%
YOLOv9 - 75.7% 83.4% 77.9% 46.7%
DINO Swin-T - - - 32.2% (Avg IoU)

OCR Performance

Model CER ↓ WER ↓
Tesseract 15.56% 30.78%
Attention-Gated CNN-BiGRU 14.85% 27.22%
Our OCR Model 7.83% 27.24%
Azura OCR 6.39% 25.97%

Receipt QA Performance

Model Precision Recall Exact Match Contains
GPT-4o 37.7% 36.4% 35.0% 29.1%
Llama3.2 (11B) 32.6% 31.3% 31.6% 25.9%
Phi3.5 28.4% 29.1% 28.8% 23.7%
Internvl2 (8B) 24.2% 23.8% 23.1% 19.4%

🚀 Getting Started

Quick Start

# Install required packages
pip install datasets transformers torch

# Load the dataset
from datasets import load_dataset

# Load Receipt QA dataset
qa_dataset = load_dataset("abdoelsayed/CORU", "qa")

# Load OCR dataset  
ocr_dataset = load_dataset("abdoelsayed/CORU", "ocr")

# Load Information Extraction dataset
ie_dataset = load_dataset("abdoelsayed/CORU", "ie")

Dataset Structure

ReceiptSense/
├── Receipt/          # Key Information Detection
│   ├── images/       # Receipt images
│   └── annotations/  # YOLO/COCO format annotations
├── OCR/             # OCR Dataset
│   ├── images/      # Text line images
│   └── labels/      # Character annotations
├── IE/              # Information Extraction
│   └── data.csv     # Structured item data
└── QA/              # Receipt Question Anshwering
    ├── images/      # Receipt images
    └── qa_pairs.json # Question-answer pairs

🔬 Applications

  • 💳 Expense Management: Automated expense tracking and categorization
  • 📦 Inventory Management: Real-time inventory updates from receipt data
  • 🏪 Retail Analytics: Customer behavior and purchasing pattern analysis
  • 🤖 Document AI: Multilingual document understanding systems
  • 📱 Mobile Apps: Receipt scanning and digitization applications

🤝 Comparison with Existing Datasets

Dataset Images Categories Languages Item IE Receipt QA Year
SROIE 1,000 4 English 2019
CORD 1,000 8 English 2019
MC-OCR 2,436 4 EN + Vietnamese 2021
UIT 2,147 4 EN + Vietnamese 2022
ReceiptSense 20,000 5 Arabic + English 2024

🏛️ Ethics and Privacy

  • All receipts collected with explicit user consent through the DISCO application
  • Comprehensive 4-step PII redaction process implemented
  • Privacy protocols strictly followed during data collection
  • Independent verification and cross-checking procedures

👥 Authors

Abdelrahman Abdallah¹, Mahmoud Abdalla², Mahmoud SalahEldin Kasem², Mohamed Mahmoud², Ibrahim Abdelhalim³, Mohamed Elkasaby⁴, Yasser Elbendary⁴, Adam Jatowt¹

¹University of Innsbruck, Innsbruck, Tyrol, Austria
²Chungbuk National University, Cheongju, Republic of Korea
³University of Louisville, Louisville, USA
⁴DISCO, Cairo, Egypt

📚 Citation

If you find ReceiptSense useful for your research, please consider citing our paper:

@article{abdallah2024receiptsense,
    title={ReceiptSense: Beyond Traditional OCR - A Dataset for Receipt Understanding},
    author={Abdelrahman Abdallah and Mahmoud Abdalla and Mahmoud SalahEldin Kasem and Mohamed Mahmoud and Ibrahim Abdelhalim and Mohamed Elkasaby and Yasser Elbendary and Adam Jatowt},
    year={2024},
    journal={ACM Conference Proceedings},
    note={Comprehensive multilingual receipt understanding dataset}
}

📄 License

This dataset is released under the MIT License. See LICENSE file for details.

🔗 Links


🌟 Star this repository if you find it helpful! 🌟

Visitors

Downloads last month
444