Lightonai's evaluation benchmark assesses vision-language models' ability to localize images within document pages using bounding boxes. The dataset was introduced in the paper 'LightOnOCR: A 1B End-to-End Multilingual Vision-Language Model for State-of-the-Art OCR'. It was last updated on the Hugging Face platform on 2026-01-23.
Use Cases
- Benchmarking model performance on image localization based on the described bounding box prediction task.
- Training models for document layout analysis based on the described figure and chart detection.
- Evaluating multilingual OCR capabilities based on the description of the associated model paper.
Strengths
- Dataset is explicitly designed as an evaluation benchmark for a specific computer vision task.
- Associated with a published research paper introducing a 1B parameter multilingual model.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- lightonai
- Collection Method
- Introduced in the paper 'LightOnOCR: A 1B End-to-End Multilingual Vision-Language Model for State-of-the-Art OCR'.
- Freshness
- Last updated 2026-01-23 13:06:52; freshness should be verified.