2023年11月-2025年8月四川成都文物高分辨率多视角纹理影像数据集

本数据集为国家重点研发计划“文物数字资源快速高效采集关键技术与设备研发”项目中“便携式文物彩色三维扫描软硬件研制”课题所构建的标准化、高质量输入数据集，为非朗伯体文物高精度数字化重建研究提供了基础数据支撑。数据内容共涵盖15个独立对象，包含3类具有次表面散射、高光等复杂光学特性的生物样本和12类涵盖陶瓷、玉器、漆器、青铜器等典型文物材质的模拟文物，可全面覆盖文物数字化重建中常见的材质类型与光学特性研究需求。数据采集工作于2023年11月至2025年8月在四川成都的实验室受控环境下完成，采集过程采用课题自研的原型扫描设备，在严格控制的均匀光照条件下对每个对象进行环绕式拍摄，获取不少于50张高分辨率彩色影像；后续经专业的色彩校正、镜头畸变校正及位姿联合优化等加工处理，同时配套基于运动恢复结构技术解算的高精度相机位姿参数，最终形成完整标准化的多视角影像序列。数据质量经综合评估，影像细节丰富、色彩还原精准，重叠度高于80%，几何一致性优良，可充分满足神经辐射场训练、多视图立体视觉重建等前沿算法的高精度应用要求，已在课题组相关算法开发验证中得到初步应用。本数据集总容量约15 GB，其中高分辨率原始影像以高保真JPG格式存储，数据格式通用易读，便于科研人员直接调用、二次开发，未来可广泛应用于文物数字化采集算法研究、文物虚拟重建技术优化等领域。

This dataset is a standardized, high-quality input dataset constructed under the subject "Development of Portable Color 3D Scanning Software and Hardware for Cultural Relics" within the National Key R&D Program of China's "Key Technologies and Equipment R&D for Rapid and Efficient Acquisition of Digital Cultural Relic Resources". It provides fundamental data support for high-precision digital reconstruction research of non-Lambertian cultural relics.

The dataset contains 15 independent objects, including 3 types of biological samples with complex optical properties such as subsurface scattering and specular highlights, and 12 types of simulated cultural relics covering typical cultural relic materials including ceramics, jade, lacquerware, bronzes and others. It fully covers the research needs for common material types and optical characteristics in digital reconstruction of cultural relics.

Data collection was carried out from November 2023 to August 2025 in a controlled laboratory environment in Chengdu, Sichuan. During the acquisition process, the self-developed prototype scanning equipment of the subject was used. Each object was captured via circular shooting under strictly controlled uniform lighting conditions, obtaining no fewer than 50 high-resolution color images. Subsequently, professional post-processing including color correction, lens distortion correction and joint pose optimization was performed. Meanwhile, high-precision camera pose parameters solved via Structure from Motion (SfM) technology were provided, ultimately forming a complete standardized multi-view image sequence.

Comprehensive quality evaluation shows that the dataset has rich image details, accurate color reproduction, an overlap rate higher than 80%, and excellent geometric consistency. It can fully meet the high-precision application requirements of cutting-edge algorithms such as Neural Radiance Fields (NeRF) training and Multi-View Stereopsis (MVS) reconstruction, and has been preliminarily applied in the development and verification of relevant algorithms by the research group.

The total capacity of this dataset is approximately 15 GB. The high-resolution original images are stored in high-fidelity JPEG format, with a universal and easy-to-read data format that allows researchers to directly call and conduct secondary development. In the future, it can be widely applied in fields such as research on digital cultural relic acquisition algorithms and optimization of virtual reconstruction technologies for cultural relics.

搜集汇总

背景与挑战

背景概述

该数据集是2023年11月至2025年8月在四川成都采集的高分辨率多视角纹理影像数据集，专为文物数字化资源快速采集的国家重点研发计划项目开发，包含15个对象，涵盖生物样本和模拟文物（如陶瓷、玉器），具有复杂光学特性。数据通过自主研发设备在受控环境下采集，每对象不少于50张高分辨率图像，并经过专业处理，支持高精度数字重建算法如NeRF训练，总容量约15GB，存储为通用JPG格式，便于研究和二次开发。