Galaxy Descriptions with Images and AI-Generated Captions

Galaxy Descriptions provides galaxy cutout images, natural-language descriptions, and embeddings for galaxies drawn from the Legacy DR10 and HSC PDR3 Wide imaging surveys. The dataset includes preprocessed RGB images, captions generated by gpt-4.1-mini, and single-sentence summaries from gpt-4.1-nano. It was created by astronolan and last updated on 2026-01-23.

Use Cases

• Train or fine-tune vision-language models for astronomy based on the paired galaxy images and AI-generated captions.
• Perform semantic search across galaxy images using the provided text or image embeddings.
• Benchmark automated captioning systems for scientific imagery against the gpt-4.1-mini generated descriptions.
• Analyze the relationship between visual galaxy features and their textual descriptions for educational or research purposes.

Strengths

• Includes data from multiple established imaging surveys: Legacy DR10 and HSC PDR3 Wide.
• Provides multiple AI-generated text descriptions (full caption and summary) per galaxy image.
• Contains precomputed text and image embeddings, potentially reducing preprocessing effort.

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

Galaxy images drawn from Legacy DR10 and HSC PDR3 Wide surveys, with text descriptions generated by AI models (gpt-4.1-mini, gpt-4.1-nano).

Freshness

Last updated 2026-01-23 03:10:04; freshness should be verified.