20250129_Fabric_defects

1. Dataset Overview

The DME Fabric Defect Detection Dataset is an industrial microscopy image collection designed for research on automated fabric defect detection and classification. The dataset captures a diverse range of real textile defects under controlled acquisition conditions and reflects practical challenges encountered in textile quality inspection environments.

The dataset is intended for use in computer vision, deep learning, and industrial inspection research, particularly for evaluating CNN-based and hybrid architectures.
Class Distribution
Class Label
Stain |Damage |Broken Thread |Holes |Non-defective
3. Fabric Types and Visual Diversity

Images were collected from multiple fabric categories to ensure diversity and generalization:

Plain fabrics

Textured fabrics

Woven fabrics

Printed fabrics

Satin fabrics

Denim fabrics

Defects vary in size, orientation, contrast, and texture, reflecting real industrial variability.
4. Image Resolution and Format

Original acquisition resolution: 1920 × 1080 pixels

Processed resolution (for model training): 128 × 128 pixels

Color format: RGB (3 channels)

File format: JPG

Downsampling was applied only during model training to ensure computational feasibility, while original high-resolution images are preserved.