Unified Text and Formula Recognition Dataset for OCR Model Training

UniRec-0.1B is a unified recognition model with 0.1 billion parameters for high-accuracy recognition of plain text and mathematical formulas in Chinese and English. The dataset supports training this model to handle structural variability and semantic understanding of mixed content. It was created by author 'topdu' and last updated on January 8, 2026.

Use Cases

• Train a model to recognize plain text features like words, lines, and paragraphs from document images.
• Fine-tune a model for recognizing single-line and multi-line mathematical formula structures.
• Develop a unified system for parsing mixed content containing both text and formula features.
• Benchmark model performance on multilingual recognition tasks using Chinese and English text features.

Strengths

• Model architecture contains 0.1 billion parameters for efficient recognition.
• Designed for high-accuracy recognition across multiple content types.

Limitations

• Dataset size, row count, and specific sample characteristics are unknown.
• The availability and quality of ground-truth annotations for training are unspecified.

Provenance

Source

Hugging Face dataset repository, author 'topdu'.

Collection Method

null

Time Range

null

Freshness

null

Geography

null