ArtiMuse: Fine-Grained Image Aesthetics with Joint Scoring and Expert Understanding

ArtiMuse provides data for fine-grained image aesthetics assessment (IAA) through joint numerical scoring and expert-level textual understanding, released by thunderbolt215 in early 2026. The dataset supports multi-modal large language models (MLLMs) by pairing visual content with detailed aesthetic critiques and quantitative scores.

Use Cases

• Training MLLMs to generate aesthetic critiques using the expert-level understanding labels
• Benchmarking image aesthetics assessment models against the joint scoring fields
• Fine-tuning vision-language models for fine-grained visual quality analysis

Strengths

• Includes expert-level textual understanding for aesthetic reasoning
• Supports joint scoring and qualitative assessment
• Associated with CVPR 2026 research

Limitations

• Lack of public documentation on total record count in the current metadata
• Potential for subjective bias inherent in expert-level aesthetic labels

Provenance

Source

CVPR 2026 publication by thunderbolt215

Freshness

Last updated February 2026.