Sikka-MoDiVers: A Curated Image Dataset of Traditional Sikka Ikat Weaving Motifs for Textile Pattern Analysis

This dataset was created under the hypothesis that image-based computational methods can effectively be used to classify and recognize traditional textile motifs, particularly Sikka ikat weaving patterns. Variations in lighting, device resolution, and motif structure are significant challenges in textile image analysis. By capturing these variations in structured dataset, we hypothesize that it will support the development of more robust and generalizable computer vision models for motif recognition and cultural heritage preservation.
The dataset presents 36 distinct Sikka ikat motif classes, images captured under three natural lighting conditions (morning, noon, and night). The images exhibit a wide range of visual properties, including different resolutions, color intensity, texture complexity, and lighting effects. Among these, 24 classes are fully verified and culturally described by experts, while the remaining 12 are pending further documentation and image collection. The dataset is suitable for training and testing machine learning and deep learning models in classification, retrieval, segmentation, and enhancement tasks.
Image were collected using DSKR cameras Nikon D5100 and various smartphone devices in real-world indoor and outdoor environments across Sikka, East Nusa Tenggara, Indonesia. No artificial lighting or digital enhancement was applied during image acquisition. Each image was manually labelled according to motif type and lighting condition and verified by expert in traditional Sikka ikat weaving. Metadata was compiled for each image, including filename, class label, resolution, capture location, lighting condition, and device used.
The dataset is organized into folders by motif class, with images named according to lighting and capture conditions. A metadata file (metadata_motif.csv) provides technical details per image, while additional files (class_description_motif.csv, motif_class_status.csv) describe the motif meanings and verification status. The data is divided into two parts, sikka_modivers_part1.zip and sikka_modivers_part2.zip. Researchers can us this dataset for developing image classification pipelines, illumination-invariant models, textile retrieval systems, and algorithms ethnographic studies involving traditional Indonesian textiles.

本数据集的构建基于以下假设：基于图像的计算方法可有效用于传统纺织纹样的分类与识别，尤其是锡卡伊卡特（ikat）编织纹样。光照条件、设备分辨率以及纹样结构的差异，是纺织图像分析中面临的核心挑战。通过结构化数据集涵盖上述各类差异，我们推测该数据集将助力开发更具鲁棒性与泛化性的计算机视觉（Computer Vision）模型，用于纹样识别与文化遗产保护。
本数据集包含36个独立的锡卡伊卡特编织纹样类别，所有图像均在三种自然光环境（早晨、正午、夜晚）下采集。这些图像涵盖丰富的视觉属性差异，包括不同分辨率、色彩强度、纹理复杂度以及光照效果。其中24个类别已由专家完成完整验证与文化释义，剩余12个类别尚待进一步文档整理与图像采集。本数据集适用于机器学习、深度学习模型在分类、检索、分割与增强任务中的训练与测试。
图像采集使用了DSKR相机、尼康D5100（Nikon D5100）相机以及多款智能手机设备，拍摄场景覆盖印尼东努沙登加拉省锡卡地区的真实室内外环境。图像采集过程未使用人工照明，亦未进行数字增强处理。每张图像均根据纹样类型与光照条件完成手动标注，并由锡卡传统伊卡特编织专家进行审核。我们为每张图像编译了元数据（metadata），内容包括文件名、类别标签、分辨率、拍摄地点、光照条件以及使用的采集设备。
本数据集按纹样类别划分文件夹，图像名称根据光照与拍摄条件命名。元数据文件（metadata_motif.csv）提供每张图像的技术细节，额外附带的文件（class_description_motif.csv、motif_class_status.csv）则阐释纹样含义与验证状态。数据集分为两个部分：sikka_modivers_part1.zip与sikka_modivers_part2.zip。研究人员可利用本数据集开发图像分类流水线、光照不变模型、纺织物检索系统，以及用于印尼传统纺织物民族志研究的相关算法。