People: Aesthetic-Train-V2 Filtered Subset for Human Faces

ROSCOSMOS provides a filtered derivative subset of the Aesthetic-Train-V2 dataset. It contains image files paired with JSON caption and metadata sidecars, packaged as WebDataset TAR shards. The subset was prepared from a people/face-oriented slice and filtered to reduce the likelihood of distributing images containing minors or age-ambiguous subjects.

Use Cases

• Train aesthetic scoring models based on image-caption pairs.
• Develop face detection or analysis models based on a people-oriented image subset.
• Fine-tune generative image models on human-centric visual data.
• Research content filtering techniques based on the automated filtering applied to the source data.

Strengths

• Derived from the established Aesthetic-Train-V2 dataset.
• Packaged in WebDataset TAR shards, a format designed for efficient large-scale data loading.
• Includes JSON sidecar files providing caption and metadata for each image.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license are unknown, which may limit suitability assessment.

Provenance

Source

Filtered derivative of zhang0jhon/Aesthetic-Train-V2.

Collection Method

Automated filtering applied to a people/face-oriented slice of the upstream dataset.

Freshness

Last updated 2026-06-09 18:26:59; freshness should be verified.