Bharat Scene Text Dataset (BSTD)

Bharat Scene Text Dataset (BSTD)

MH Specific

Large-scale scene text dataset for 11 Indian languages plus English, sourced from Wikimedia images of Indian signboards and street scenes. Includes 5,113 Marathi word annotations with polygon bounding boxes

Build a Devanagari scene text recognition system for reading Marathi shop signs and street nameplates in urban Maharashtra.
HomepageGitHub

Quick Start

# Bharat Scene Text Dataset
import json
with open('bstd_annotations.json') as f:
    data = json.load(f)
print(f'Total images: {len(data)}')
for img in list(data.values())[:3]:
    print(f"Text regions: {len(img.get('annotations', []))}")
Modality
Image (scene text)
Size
6,582 images; 106K+ word instances
License
Format
PNG/JPEG
Language
mr
Update Frequency
static
Organization
IIIT Hyderabad

Schema

FieldTypeDescription
imageimageStreet scene image containing text
bounding_boxeslist[object]Text region bounding box coordinates
textstringRecognized text content
scriptstringScript type (Devanagari, Latin, etc.)

Build With This

Create a real-time Marathi signboard translator app that captures and translates Devanagari text from camera feeds
Develop an automated address reader for delivery services that extracts Marathi text from building photographs
Build a heritage site signage digitizer that reads and catalogs inscriptions from Maharashtra historical monuments

AI Use Cases

Scene text detectionscript identificationend-to-end OCR
Last verified: 2026-03-07