VLM-150M: A Large-Scale Recaptioned Image-Text Dataset

VLM-150M is a large-scale image-text dataset recaptioned using an SFT-enhanced Qwen2VL model to improve the alignment and detail of textual descriptions. The dataset was created by zhixiangwei and was last updated on July 28, 2025. Its repository is hosted at https://zxwei.site/hqclip/.

Use Cases

• Training vision-language models based on enhanced image-text alignment.
• Fine-tuning captioning models based on detailed textual descriptions.
• Benchmarking model performance on recaptioned image-text data.
• Conducting research on the impact of caption quality on multimodal learning.

Strengths

• Dataset name indicates a scale of 150 million items.
• Descriptions have been enhanced for detail and alignment using a specialized Qwen2VL 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

huggingface

Collection Method

Recaptioned from an original large-scale image-text dataset using an SFT-enhanced Qwen2VL model.

Freshness

Last updated 2025-07-28 06:47:50; freshness should be verified.