Ever wondered if artificial intelligence could unlock the mysteries of a lost civilization? Now's your chance! We're thrilled to announce the Great Daxinzhuang Pottery Puzzle AI Challenge on Kaggle. This unique competition fuses cutting-edge machine learning with the enigmatic world of the Shang Dynasty (c. 1600–1046 BC), offering a chance to reshape how we uncover the past.

Imagine thousands of broken pottery sherds, each a tiny piece of a forgotten story, brimming with the echoes of a powerful ancient civilization. For archaeologists, piecing these fragments back together is like solving a giant, intricate jigsaw puzzle – a painstaking process that can take months, even years, with often incomplete results. The sheer volume of unearthed fragments presents a monumental challenge to manual methods.

Our competition centers around a truly remarkable archaeological site: Daxinzhuang, a crucial Shang Dynasty center in eastern China. Specifically, you'll be working with nearly 20,000 pottery fragments unearthed from H690, a fascinating ancient storage pit that later became a refuse dump. This pit, a circular structure with a diameter of 5.2 meters and a flat wooden-floored base, even features a central post-hole (0.2 meters in diameter, 1.3 meters deep) indicating its original function as a granary. The pit's deep, 4-meter deposit, divided into 14 layers, yielded not only pottery but also animal bones and even remnants of gold foil, highlighting its exceptional importance. These artifacts hold invaluable clues about the Shang Dynasty's expansion and its intricate interactions with local cultures.   

However, despite meticulous manual classification and matching, the information gleaned from H690's vast pottery collection remains limited. More systematic and highly efficient methods for classifying and matching pottery sherds are urgently needed. This is where the transformative power of artificial intelligence, particularly machine learning techniques, proves indispensable.

This isn't just a coding challenge; it's an opportunity to gain access to an exclusive dataset of authentic archaeological artifacts, solve a real-world archaeological dilemma, and help advance the application of AI in cultural heritage preservation and computational archaeology. 

The core mission of this challenge is to develop powerful and innovative artificial intelligence models capable of solving the following key problems: 

(1) Fracture Edge Matching and Assembly: To reassemble as many excavated pottery sherds as possible. This involves developing algorithms for extracting and matching fracture edge features to compute the geometric fit of fracture edges, thereby enabling the automatic inference of pottery sherd adjacency relationships and laying the foundation for the restoration of original artifacts. Examples of successful assemblies can be found in the submission sample.

(2) Vessel 3D Reconstruction: To restore assembled pottery sherds as closely as possible to their original vessel forms.

To achieve these goals, we recommend that participants first perform intelligent classification of pottery sherds based on characteristics such as clay texture and color. During the matching process, comprehensive consideration should be given to both visual features from the images and metadata such as clay type, color, decorative patterns, and artifact part, to aid in the matching. We encourage participants to explore various innovative methods, including but not limited to: 
A. Advanced image segmentation and feature extraction techniques: to precisely delineate sherd boundaries and capture subtle texture and decorative pattern details.
B. Deep Metric Learning: to learn the similarity of pottery sherd edge features.
C. Graph Neural Networks (GNNs): to model relationships between pottery sherds and infer potential connections.
D. Multimodal Models: to combine visual information from pottery sherd images with structured metadata for more robust classification and matching.

This competition encourages exploratory rather than restrictive problem-solving approaches, and diverse and creative solutions are welcome.

We call on data scientists, AI innovators, archaeologists, and curious minds worldwide to harness the power of machine learning. Your mission: develop groundbreaking AI models to automatically identify and match these ancient pottery sherds, transforming a laborious process into a triumph of technology. 

This is far more than just a coding challenge; it's an unparalleled opportunity to access an exclusive and challenging dataset of authentic and irreplaceable archaeological artifacts, solve a real-world puzzle, and contribute to the advancement of AI in cultural heritage preservation and computational archaeology. Your intelligent solutions will not merely be lines of code; Your algorithms will weave together fragments of history, connecting ancient stories to modern innovation.

**For details regarding the datasets, please refer to The Great Daxinzhuang Pottery Puzzle Challenge - Description.pdf.**

Join this groundbreaking interdisciplinary adventure. Let’s wield AI to piece together the Shang Dynasty’s legacy, one sherd at a time, to unlock the grandeur of the Shang Dynasty!

This competition is organized by the Institute for Advanced Study at Beijing Normal University-Hong Kong Baptist University and the School of Archaeology (Institute of Cultural Heritage) at Shandong University.

你是否想过，人工智能可以如何揭开失落文明的秘密？现在，这个时刻来了！我们隆重推出“大辛庄陶片拼合AI挑战赛” - 一场融合尖端机器学习技术与商代考古的国际赛事，诚邀您一同开启科技赋能历史的新篇章。
	
想象成千上万的陶器碎片，每一片都铭刻着三千多年前的故事。考古学家们面临着一个艰巨的挑战：将这些碎片拼合成完整的器物，仿佛在解一个复杂且残缺的拼图。这一过程耗时数月甚至数年，成果却常常不尽如人意。
	
济南大辛庄遗址，商朝东方的重要聚落，为我们提供了这一挑战的舞台。2003年发掘的H690是一座直径5.2米、深4米的圆形储藏坑，后转为垃圾坑，内含14层堆积，出土了近20,000片陶器碎片、动物骨骼及珍贵的金箔残片。显示了其重要性与特殊性。因此，对H690出土陶器进行系统深入的研究，对于了解商王朝在大辛庄遗址的活动以及商族与本地东夷族的复杂互动具有极其重要的意义 。
	
面对这么多陶片，怎么才能又快又好地把它们分类、拼好，从中找出更多的历史信息呢？这正是考古学界急需解决的难题。人工智能和深度学习技术无疑是当下最合适的方法之一，它们是处理大数据和识别模式的“超级武器” 。

本次挑战的核心使命是开发强大且创新的人工智能模型，能够解决以下关键问题：

(1) 断茬匹配拼合： 将出土的陶片尽可能多地重新拼对起来 。通过开发断茬边缘特征提取与匹配算法，对断茬边缘形状几何匹配度进行计算，实现陶片邻接关系的自动推断，从而为原始器物复原奠定基础 。拼合成功的案例参见提交样例 。

(2)器形三维重建： 将拼合起来的陶片尽可能按照原始器形复原 。

为实现这些目标，我们建议参赛者首先对陶片进行智能分类，将出土的陶片按照陶质、陶色等进行分类，在拼对的过程中综合考虑陶片图像特征与元数据，如陶质、陶色、纹饰、器型部位等辅助拼对 。我们鼓励参赛者探索各种创新方法，包括但不限于 ：

先进的图像分割和特征提取技术：精确描绘陶片边界，捕捉细微的纹理和纹饰细节 。

深度度量学习 (Deep Metric Learning) 来学习陶片边缘特征的相似度 。

图神经网络 (Graph Neural Networks) 建模陶片之间的关系，推断潜在连接关系 。

多模态模型 (Multimodal Models)：结合陶片图像的视觉信息与结构化元数据，以实现更稳健的分类和匹配 。

本次竞赛鼓励采用探索性而非限制性的问题解决方法，欢迎多样化和创造性的解决方案 。
	
我们诚挚地邀请全球的数据科学家、AI开发者、考古学家以及所有对文化遗产保护有浓厚兴趣的朋友们，一起来用先进的机器学习技术，开发出能够自动识别并精确匹配商代陶片连接关系的创新模型 。这不仅仅是一场技术实力的巅峰对决，更是一次推动科技深度服务人文的实践！ 你将有机会接触并分析真实、充满挑战性的独一无二的考古数据；解决真正的考古难题，直接帮助考古学家高效地整理和分析珍贵文物，为历史研究提供前所未有的强大支持；拓展AI的应用范围，推动人工智能在考古以及文化遗产保护领域的深入用，为“计算考古学”这一新兴且充满潜力的领域注入蓬勃生机；搭建连接历史与未来的桥梁，你的智慧结晶，将不仅仅是冰冷的代码，它会成为连接尘封往事和现代科技的宏伟纽带 。

**关于竞赛数据集等详情，请参考The Great Daxinzhuang Pottery Puzzle Challenge - Description.pdf**
	
期待你的加入，让我们一起揭开大辛庄陶片背后千年的秘密，用AI的力量重现商代的辉煌！

本次竞赛主办单位：北师香港浸会大学高等研究院与山东大学考古学院（文化遗产研究院）