Odia OCR Merged: 192,000+ Handwritten and Digital Text Samples

192,000+ Odia OCR samples merged from three public sources by shantipriya, updated in February 2026. The collection integrates handwritten and digital text images to support character recognition and script analysis for the Odia language.

Use Cases

• Training handwritten character recognition models using the 182,152 samples from the tell2jyoti subset
• Fine-tuning multilingual OCR systems for Indic scripts using the indic-mozhi-ocr subset
• Benchmarking Odia script detection across varied digital and manual writing styles

Strengths

• 192,000+ total samples
• 182,152 handwritten samples
• Consolidates three distinct source datasets into a single repository

Limitations

• Significant class imbalance with 94% of data originating from a single handwritten source
• Potential label inconsistency across merged sources
• Extremely small sample size (64) for the Odia-lipi-ocr-data subset

Provenance

Source

Merged from OdiaGenAIOCR, tell2jyoti, and darknight054

Collection Method

Merged from existing public datasets

Freshness

Last updated February 2026

Geography

India