Scholarlink

<p>Bronze inscriptions (BI), engraved on ritual vessels, constitute a crucial stage of early Chinese writing and provide indispensable evidence for archeological and historical studies. However, automatic BI recognition is challenging due to visual degradation, cross-domain variability among photographs, rubbings, and tracings, and an extremely long-tailed character distribution. To address these challenges, we curate a large-scale BI dataset comprising 22,454 full-page images and 198,598 annotated characters spanning 6658 unique categories, enabling robust cross-domain evaluation. Building on this resource, we develop a two-stage detection-recognition pipeline. To handle heterogeneous domains and rare classes, we equip the pipeline with LadderMoE, which augments a pretrained CLIP encoder with ladder-style MoE adapters for dynamic expert specialization and enhanced robustness. Comprehensive experiments demonstrate that our method substantially outperforms the state-of-the-art scene text recognition baselines, achieving superior accuracy across head, mid, and tail categories as well as all acquisition modalities, establishing a strong foundation for downstream archeological analysis.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Ladder-side mixture of experts adapters for bronze inscription recognition

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月15日

查看全文
  • Rixin Zhou,
  • Peiqiang Qiu,
  • Qian Zhang,
  • Chuntao Li,
  • Xi Yang

摘要

Bronze inscriptions (BI), engraved on ritual vessels, constitute a crucial stage of early Chinese writing and provide indispensable evidence for archeological and historical studies. However, automatic BI recognition is challenging due to visual degradation, cross-domain variability among photographs, rubbings, and tracings, and an extremely long-tailed character distribution. To address these challenges, we curate a large-scale BI dataset comprising 22,454 full-page images and 198,598 annotated characters spanning 6658 unique categories, enabling robust cross-domain evaluation. Building on this resource, we develop a two-stage detection-recognition pipeline. To handle heterogeneous domains and rare classes, we equip the pipeline with LadderMoE, which augments a pretrained CLIP encoder with ladder-style MoE adapters for dynamic expert specialization and enhanced robustness. Comprehensive experiments demonstrate that our method substantially outperforms the state-of-the-art scene text recognition baselines, achieving superior accuracy across head, mid, and tail categories as well as all acquisition modalities, establishing a strong foundation for downstream archeological analysis.