HCSU introduces a benchmark for fine-grained historical calligraphy style understanding. It contains 39,307 character images from 49 historically prominent calligraphers spanning 10 dynasties, while explicitly separating authentic ink manuscripts (Tie) from stone rubbings (Bei). The dataset also provides hierarchical, expert-written aesthetic descriptions for style discrimination and interpretable aesthetic reasoning. According to the paper’s abstract, current LVLMs achieve non-trivial results but remain sensitive to script-level, textual, and source-specific cues, and often fail to ground aesthetic judgments in detailed brushwork evidence.
No heat snapshots are available in the last 24 hours.