BLIP3o-Pretrain-Long-Caption: 27 Million Images with Long Synthetic Captions

A collection of 27 million images, each paired with a long caption generated by the Qwen2.5-VL-7B-Instruct model. The dataset was created by the BLIP3o organization and published on Hugging Face in June 2025. It is intended for pretraining vision-language models.

Use Cases

• Pretraining vision-language models based on the large-scale image-caption pairs.
• Improving long-form image caption generation based on the ~120-token captions.
• Benchmarking synthetic caption quality based on outputs from the Qwen2.5-VL model.
• Training models for detailed visual understanding based on descriptive captions.

Strengths

• Contains 27 million images, providing a large-scale resource.
• Each image has a long caption of approximately 120 tokens, offering detailed descriptions.
• Captions are generated by a specific, named model (Qwen2.5-VL-7B-Instruct), providing traceability.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the source images and caption generation model.

Provenance

Source

BLIP3o organization via Hugging Face.

Collection Method

Images paired with captions synthetically generated by the Qwen2.5-VL-7B-Instruct model.

Time Range

null

Freshness

Last updated 2025-06-26 17:54:21.

Geography

null