FaceCaptionHQ-4M: 4 Million Facial Image-Text Pairs

FaceCaptionHQ-4M is a dataset containing approximately 4 million facial image-text pairs. It was created by OpenFace-CQUPT and was last updated on 2025-06-09. The dataset is a cleaned subset derived from the larger FaceCaption-15M dataset.

Use Cases

• Training facial image captioning models based on the described image-text pairs.
• Fine-tuning vision-language models for tasks involving human faces.
• Researching bias and representation in facial image datasets.
• Developing models for attribute-based facial image retrieval using textual descriptions.

Strengths

• Contains approximately 4 million data points, indicating a substantial scale.
• Is a cleaned version of a larger source dataset, which suggests some quality control.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Access requires a separate application and agreement process for the original source images.
• Row count is approximate, and specific data quality metrics are not provided in the description.

Provenance

Source

OpenFace-CQUPT via Hugging Face.

Collection Method

Cleaned from the FaceCaption-15M dataset.

Freshness

Last updated 2025-06-09 13:21:36.