陶器
工程类
视觉艺术
考古
艺术
线条图
窗口(计算)
组合(考古学)
多色
历史
作者
Quan Zheng,Yuting Mei,Xinfeng Chang,Yongdong Zhang,Miaole Hou
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-07-01
卷期号:: 1-3
标识
DOI:10.1109/icmew69720.2026.11720793
摘要
Reassembling broken pottery is a labor-intensive yet demanding task. It often requires comparison of fragments, careful spatial alignment, and expert interpretation. This makes pottery restoration difficult to scale in archaeology and cultural heritage preservation. In this demo, we present SherdAgent, a novel interactive multimodal AI agent for pottery sherd reassembly and restoration, leveraging recent advanced AI models. The system brings fragment understanding, 3D reassembly, and digital restoration into one unified interface. It is designed not only as a technical tool, but also as a human-in-the-loop platform for archaeologists, conservators, and digital heritage researchers. Given images of excavated pottery sherds, SherdAgent first analyzes visual and semantic cues to group related fragments. It then lifts the grouped sherds into 3D using a pre-trained image-to-3D model, and supports spatial reassembly by an off-the-shelf 3D reassembly model. Finally, the system restores missing regions and generates a coherent digital pottery model with a tailored generative AI model. Through natural interaction, users can inspect intermediate results, refine the reassembly process, and obtain complete 3D restorations more efficiently. Our live demo highlights how recent multimodal AI can support cultural heritage workflows in an intuitive and creative way. It shows the potential of AI agents to significantly reduce manual effort and improve restoration efficiency, as well as open new possibilities for archaeological analysis, digital preservation, and virtual exhibition.
科研通智能强力驱动
Strongly Powered by AbleSci AI