Fabric Defects Dataset

The data collected for this project were from :
● Fabric Defect Dataset from Kaggle (Ranathunga, 2020)
● Fabric Stain Dataset from Kaggle (Pathirana, 2020).
● Aitex fabric image database (Silvestre-Blanes et al., 2019)
● Dataset from the author of the literature review paper (Peng, 2020): only the non-defect images from the dataset.
Aitex fabric image database
This textile fabric dataset consisted of 245 images of 7 different fabrics with image sizes of 4096×256 pixels. There were 140 defect-free images, 20 for each type of fabric. In each of the defected category, there were 105 images (Silvestre-Blanes et al., 2019).
Fabric Stain Dataset
This was taken from Kaggle (Pathirana, 2020). The dataset was built as a part of the fabric defect detection project of the Intelligence Lab of University of Moratuwa, Sri Lanka. The dataset consisted of images with resolution of 1920x1080 or 1080x1920. It consisted of 398 defected images with different types of stains and 68 defect free images.
Fabric Defect Dataset
The dataset was also taken from Kaggle and it was supplied by the Intelligence Lab, Department of computer science and Engineering, University of Moratuwa (Ranathunga, 2020). This dataset consisted of 3 classes of defective images namely horizontal, vertical and holes along with 3 mask images for each defective image sample. The folder named 'captured' consisted of raw images. Images have a size of 640x360.
Dataset from the author of the literature review paper
Authors of the research papers in the literature review were contacted through emails for the dataset. With the consent of the author of the paper “Automatic fabric defect detection method using pran-net” (Peng, 2020) named Troy Peng, the annotated data was used (Peng et. al, 2020).
After collecting the images from all these sources, the distribution of the images is as given below:
● Defect free 1,666
● Defected 1,073
Defected hole 281
Defected horizontal 136
Defected lines images 157
Defected stain images 398
Defected verticle images 101
Total Images = 2,739

本项目所采集的数据集来源如下：
● Kaggle平台织物缺陷数据集（Fabric Defect Dataset，Ranathunga, 2020）
● Kaggle平台织物污渍数据集（Fabric Stain Dataset，Pathirana, 2020）
● Aitex织物图像数据库（Silvestre-Blanes et al., 2019）
● 该综述论文作者提供的数据集（Peng, 2020）：仅使用其中的无缺陷图像。

### Aitex织物图像数据库
该纺织织物数据集包含7种不同织物的245张图像，图像分辨率为4096×256像素。其中无缺陷图像共140张，每种织物对应20张；各类缺陷图像总计105张（Silvestre-Blanes et al., 2019）。

### 织物污渍数据集（Fabric Stain Dataset）
该数据集源自Kaggle平台（Pathirana, 2020），由斯里兰卡莫拉图瓦大学智能实验室的织物缺陷检测项目构建。数据集包含分辨率为1920×1080或1080×1920的图像，其中包含398张含各类污渍的缺陷图像，以及68张无缺陷图像。

### 织物缺陷数据集（Fabric Defect Dataset）
该数据集同样取自Kaggle平台，由莫拉图瓦大学计算机科学与工程系智能实验室提供（Ranathunga, 2020）。本数据集包含3类缺陷图像：横向缺陷、纵向缺陷与孔洞缺陷，且每张缺陷图像对应3张掩码图像（mask images）。名为"captured"的文件夹存储原始图像，图像分辨率为640×360。

### 综述论文作者提供的数据集
本团队通过邮件联系了该综述涉及的研究论文作者以获取数据集。经论文《Automatic fabric defect detection method using pran-net》（Peng, 2020）的作者Troy Peng同意后，使用了其团队标注的相关数据（Peng et. al, 2020）。

完成所有数据源采集后，各类图像的分布如下：
● 无缺陷图像：1666张
● 缺陷图像：1073张
- 孔洞缺陷图像：281张
- 横向缺陷图像：136张
- 线条缺陷图像：157张
- 污渍缺陷图像：398张
- 纵向缺陷图像：101张
● 总图像数：2739张

搜集汇总

背景与挑战

背景概述

Fabric Defects Dataset的版本1已被移除，当前可用版本为Version 324 October 2024，但缺乏具体内容描述。