Corvus-OCR-Caption-Mix: 229,000 Image-Caption Pairs for OCR and Long-Form Captioning

Corvus-OCR-Caption-Mix is a high-quality, compact multimodal dataset containing over 229,000 image-caption pairs. It was created by author zhangshengchoa and is derived from the larger BLIP3o/BLIP3o-Pretrain-Long-Caption dataset. The dataset was last updated on the platform in April 2026.

Use Cases

• Training image-to-text models based on the described long-form captions.
• Evaluating OCR performance on images containing text, based on the dataset's focus on OCR-rich documents.
• Fine-tuning multimodal models for captioning a variety of image types as mentioned in the description.

Strengths

• Contains over 229,000 image-caption pairs, providing a substantial training corpus.
• Designed for high quality and a compact size, optimized from a larger source dataset.
• Focuses on long-form captions and OCR tasks across a variety of image types.

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

Derived and optimized from the BLIP3o/BLIP3o-Pretrain-Long-Caption dataset.

Collection Method

Derived and optimized from a larger source dataset; specific gathering method is not detailed.

Time Range

null

Freshness

Last updated 2026-04-01 04:54:02; freshness should be verified.

Geography

null