Research map / Vision-language and multimodal / Multimodal LLMs
HCSU: A Dataset and Benchmark for Fine-Grained Historical Calligraphy Style Understanding
Yinsheng Yao, Yan Liu, Chen Ye
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use raw Wild images instead of processed Origin images (ablation) | Qwen3-VL-235B-Instruct · HCSU discrimination | accuracy: 29.95 -> 34.22 | Improved |
| Use expert descriptions instead of random ones (Bei) (ablation) | Qwen3-VL-235B-Instruct · HCSU discrimination (Bei) | accuracy: 30.93 -> 34.10 | Improved |
| Use expert descriptions instead of GPT-generated ones (Tie) (ablation) | Qwen3-VL-235B-Instruct · HCSU discrimination (Tie) | accuracy: 26.74 -> 26.40 | No clear effect |
| Use expert descriptions instead of random ones (Tie) (ablation) | Qwen3-VL-235B-Instruct · HCSU discrimination (Tie) | accuracy: 24.84 -> 26.40 | Improved |
| Scale Qwen3-VL from 30B to 235B | Qwen3-VL · HCSU discrimination | avg accuracy: 21.81 -> 30.25 | Improved |
| Scale Qwen2.5-VL from 7B to 72B | Qwen2.5-VL · HCSU discrimination | avg accuracy: 20.25 -> 24.93 | Improved |
| Evaluate on ink manuscripts (Tie) instead of stone rubbings (Bei) | Qwen3-VL-235B-Instruct · HCSU discrimination | accuracy: 34.10 -> 26.40 | Hurt |
| Resize Wild images instead of raw (ablation) | Qwen3-VL-235B-Instruct · HCSU discrimination | accuracy: 34.22 -> 31.40 | Hurt |
| Image condition (raw vs resized vs processed) (ablation) | Qwen3-VL-235B-Instruct · HCSU generation | BERTScore: 6.94-6.96 | No clear effect |
| Go from 1-shot to 2-shot in aesthetic description prompts (ablation) | Qwen3-VL-235B probe · HCSU description generation | Terminology: +0.33 | Improved |
| Best discrimination model vs best generation model | Doubao-1.5-Vision-Pro vs GPT-5.2 · HCSU | discrimination vs richness: 37.29 vs 34.58 acc; Rich. -1.46 vs +0.82 (Bei) | Mixed |
Closest papers by meaning
- Towards Automated Chinese Ancient Character Restoration: A Diffusion-Based Method with a New Dataset (AAAI 2024 · similarity 0.86)
- OBI-Bench: Can LMMs Aid in Study of Ancient Script on Oracle Bones? (ICLR 2025 · similarity 0.85)
- Gender Artifacts from Art History to Text-to-Image Generation (NeurIPS 2026 · similarity 0.85)
- ASCIIEval: Benchmarking Models' Visual Perception in Text Strings via ASCII Art (ICLR 2026 · similarity 0.84)
- Beyond Aesthetics: Cultural Competence in Text-to-Image Models (NeurIPS 2024 · similarity 0.84)
- APDDv2: Aesthetics of Paintings and Drawings Dataset with Artist Labeled Scores and Comments (NeurIPS 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
