Document Rotation Benchmark for OCR Orientation Detection

The ORB Rotation Benchmark is a multi-domain dataset for evaluating document orientation detection models and their impact on Optical Character Recognition performance. It introduces controlled rotation settings including 4-class, 8-class, and 12-class orientation tasks.

Use Cases

• Evaluate document orientation detection models on controlled 4-class, 8-class, and 12-class rotation settings.
• Assess the impact of corrected orientation on downstream Optical Character Recognition performance.
• Benchmark multi-domain document analysis models using structured rotation tasks.

Strengths

• Designed for a specific benchmark task from a published research paper (arXiv:2511.04161).
• Provides structured multi-domain data for controlled evaluation of orientation detection.

Limitations

• Dataset size, specific column structure, and sample data are unavailable for assessment.
• The scope and diversity of document domains within the dataset are not detailed.

Provenance

Source

krutrim-ai-labs on Hugging Face.

Collection Method

Created for the research paper 'Seeing Straight: Document Orientation Detection for Efficient OCR'.

Freshness

Last updated on 2026-02-24.