You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

🏛️ Indic-CIB: Citizen Information Board Dataset

A Benchmark for Multilingual Indic Scene Text & Tabular Financial Extraction

License: Apache 2.0 Dataset Type Scripts Domain

Bridging the gap between rural public governance signage and open-source Document Intelligence.


📌 Dataset Summary

Under major Indian public works initiatives—such as the Mahatma Gandhi National Rural Employment Guarantee Scheme (MGNREGA) and the **Pradhan Mantri Gram Sadak Yojana (PMGSY)**—local governing bodies (Gram Panchayats) are mandated to erect physical Citizen Information Boards (नागरिक सूचना पट्ट / CIB) at public asset sites.

These outdoor installations serve as the bedrock of grassroots public auditing, displaying critical civic and financial data:

  • Asset Description: Details of the physical asset (e.g., roads, farm ponds, percolation tanks, check dams).
  • Financial Breakdowns: Sanctioned expenditure, skilled material allocations, and unskilled manual labor funds.
  • Administrative Metadata: Unique work identification codes, implementing agencies, and execution timelines.

Extracting structured data from real-world CIBs presents unique computer vision challenges: outdoor weathering, hand-painted lettering, complex tabular grids, and regional Indic typography. The Indic-CIB Dataset provides procedurally generated, ground-truth-annotated boards designed to benchmark and train Vision-Language Models (VLMs) and OCR layout pipelines for automated civic audits.


🌐 Supported Scripts & Languages

This benchmark currently spans two major linguistic families of India, covering four prominent regional languages:

Language Family Language Script Region / Primary Context
Indo-Aryan Hindi (हिन्दी) Devanagari Northern & Central Gram Panchayats
Indo-Aryan Marathi (मराठी) Devanagari Western India (Maharashtra MGNREGS / Jalyukt Shivar)
Dravidian Tamil (தமிழ்) Tamil Southern India (Tamil Nadu Rural Works)
Dravidian Telugu (తెలుగు) Telugu Southern India (Andhra Pradesh & Telangana Panchayats)

🗂️ Dataset Structure & Fields

Each entry in the dataset pairs a high-resolution synthesized board image with standardized key-value metadata and raw tabular Markdown.

Field Definitions

  • file_name (string): Relative file path to the board image within the repository (train/cib_...jpg).
  • language (string): ISO-style language identifier (hindi, marathi, tamil, telugu).
  • script (string): Writing system classification (devanagari or dravidian).
  • work_name (string): Name and description of the civil works project in the native regional script.
  • work_id (string): Official alphanumeric identification code assigned to the project.
  • sanctioned_amount_inr (float): Total approved capital expenditure in Indian National Rupees (₹).
  • labor_cost_inr (float): Component allocated for unskilled labor and wage compensation.
  • material_cost_inr (float): Component reserved for raw materials and technical equipment.
  • completion_date (string): Scheduled or recorded completion timestamp (DD/MM/YYYY).
  • table_markdown (string): Clean Markdown table representation of the board's core financial ledger.

Annotation Preview (metadata.jsonl)

{
  "file_name": "train/cib_marathi_0001.jpg",
  "language": "marathi",
  "script": "devanagari",
  "work_name": "पांदण रस्ता खडीकरण",
  "work_id": "IN-MA-482019",
  "sanctioned_amount_inr": 450000.0,
  "labor_cost_inr": 270000.0,
  "material_cost_inr": 180000.0,
  "completion_date": "24/11/2024",
  "table_markdown": "| Field | Value |\n|---|---|\n| Work | पांदण रस्ता खडीकरण |\n| Sanctioned | ₹4,50,000.00 |\n| Labor | ₹2,70,000.00 |\n| Material | ₹1,80,000.00 |"
}

📜 License

This project is licensed under the Apache License 2.0.


📑 Citation

If you use this dataset or tool in your research or project, please cite it using the following format:

@misc{indic_cib_dataset,
  title={Indic-CIB: Synthetic Benchmark for Citizen Information Board OCR and Tabular Extraction},
  author={Patil, Rohit Manikrao},
  year={2026},
  publisher={Hugging Face},
  howpublished={\url{[https://huggingface.co/datasets/Litap007/indic-citizen-information-board](https://huggingface.co/datasets/Litap007/indic-citizen-information-board)}}
}

Downloads last month
63