534 East Nusa Tenggara Traditional Textile Motif Image Dataset
Description
This dataset presents a structured collection of digital images of traditional textile motifs documented in East Nusa Tenggara (Nusa Tenggara Timur/NTT), Indonesia, particularly from the Sumba and Flores region. It was developed as part of a cultural heritage documentation initiative to preserve the visual characteristics and contextual information associated with traditional Indonesian textiles while providing reusable data for image processing and computer vision research. The dataset contains original textile images, augmented images, and structured metadata. Each original image is associated with five augmentation variants: rotation, horizontal flip, vertical flip, brightness adjustment, and grayscale transformation. The relationship between the original and augmented images is explicitly maintained in the metadata, enabling researchers to distinguish field-documented source images from computationally generated variants. Each record contains metadata describing the textile type, motif name, motif category, province, district/city, philosophical meaning, production technique, and documentation date. The collection represents diverse visual categories, including geometric, fauna, flora, human/anthropomorphic, and combined motifs. It also documents several traditional weaving techniques, including tenun ikat, tenun ikat lungsi, and tenun ikat pahikung. The dataset is designed for interdisciplinary research in computer vision, image processing, machine learning, deep learning, digital cultural heritage, and traditional textile studies. Potential applications include motif classification, visual feature extraction, similarity analysis, clustering, content-based image retrieval, transfer learning, and development of automated traditional textile recognition systems. Beyond visual recognition, the inclusion of cultural and geographical metadata enables researchers to connect computational analysis with the cultural context of each textile. The dataset can therefore serve as a foundation for developing AI-based approaches to the documentation, identification, and preservation of Indonesian traditional textile motifs. When developing machine-learning models, augmented images originating from the same original image should remain within the same data partition to prevent data leakage between training, validation, and testing sets.
Files
Steps to reproduce
The NTT Traditional Textile Motif Image Dataset was developed through a structured process of documentation, image selection, motif classification, augmentation, metadata compilation, and validation. 1. Textile Documentation Traditional woven textiles from East Nusa Tenggara (NTT) were documented from several textile-producing areas, particularly across Sumba and Flores. Each textile was photographed to capture its motif structure, visual composition, color, and weaving characteristics. 2. Image Selection The collected photographs were reviewed, and representative images were selected based on motif visibility, visual clarity, and suitability for digital documentation and image-processing applications. Each selected original image was maintained as an individual dataset record. 3. Motif Identification and Categorization Each image was examined to identify the motif name when a recognized designation was available, including motifs such as Mamoli, Patola, and Ketupat. Where a specific traditional name could not be confirmed, a descriptive visual label was retained rather than assigning an unsupported designation. Motifs were categorized as geometric, flora, fauna, figurative, and imaginary/symbolic (khayalan), with multiple categories assigned when appropriate. 4. Textile and Technique Classification Each record was classified according to the available textile and production information. The dataset consists of traditional woven textiles (tenun) using techniques including tenun ikat, tenun ikat lungsi (warp ikat), and tenun ikat pahikung. Broader classifications were retained when a more specific technique could not be sufficiently established. 5. Cultural Metadata Compilation Cultural and philosophical descriptions were recorded based on available validation information. These include associations with ancestors, fertility, prosperity, protection, strength, courage, social status, balance, ceremonial practices, and relationships between humans and nature. Where a specific meaning could not be sufficiently verified, broader descriptions were retained to avoid unsupported cultural claims. 6. Image Augmentation Each original image was supplemented with five standardized augmented versions: rotation, horizontal flip, vertical flip, brightness adjustment, and grayscale conversion. Thus, each record links one original image with five derived images, while preserving the original separately from computational transformations. 7. Metadata Structuring and Validation Each record was organized in JSON format containing the original file path, augmented files, textile type, motif name or description, motif category, province, regency/locality, philosophical meaning, production technique, and documentation date. The final records were reviewed to ensure consistency between images and metadata. Information that could not be sufficiently verified was not converted into definitive cultural claims.
Institutions
- State University of GorontaloGorontalo, Gorontalo