CompoVista uses AI to analyze Chinese paintings at scale
New AI system CompoVista maps compositional patterns in 10,000+ Chinese paintings...
Researchers from multiple Chinese universities have developed CompoVista, a visual analytics system that applies AI to systematically analyze compositional patterns in Traditional Chinese Paintings (TCPs) at scale.
CompoVista introduces a Composition Graph representation that models paintings through four layers: entities (objects/elements), relations (spatial/structural links), void space (negative space distribution), and context (cultural/aesthetic framing). This structured representation enables art historians to construct and refine painting cohorts through visual and context queries, inspect entity distributions and relations at cohort level, compare compositional differences across groups, and trace aggregate patterns back to individual paintings. The system was evaluated through a task-based user study with 12 domain participants, two case studies, and expert interviews, demonstrating its ability to support composition-oriented cohort construction, pattern discovery, iterative refinement, and evidence inspection—capabilities that were previously difficult with qualitative, interpretation-driven methods.
- CompoVista models paintings as 4-layer Composition Graphs (entities, relations, void space, context) for systematic analysis
- Enables art historians to query cohorts, compare patterns, and trace findings across 10,000+ paintings
- Evaluated with 12 experts; improves on traditional qualitative, interpretation-driven methods
Why It Matters
Revolutionizes art history research by enabling data-driven analysis of compositional patterns at scale for the first time.