A summary of open source industrial defect datasets, which are being continuously updated (28 have been updated)

Reprint: A summary of open source industrial defect datasets, which are being continuously updated (28 have been updated) - Zhihu welcomes everyone to pay attention to my public account: Moment AI This article currently summarizes 28 common open source industrial defect datasets, which are being continuously updated (welcome Please leave a message to add, and jointly build an article that provides convenience for everyone) The official link of the Northeastern University hot-rolled strip surface defect dataset: Vision-based SIS… https://zhuanlan.zhihu.com/p/195699093

This article currently summarizes 28 common open source industrial defect data sets, which are being continuously updated

(You are welcome to leave a message to add, and jointly build an article that provides convenience for everyone)

Northeastern University Hot Rolled Strip Surface Defect Dataset

Official link: Vision-based SIS for steel

This data set was collected by several teachers including Song Kechen from Northeastern University. It contains three types of data. Sometimes the official website cannot be opened. I saved all three defect data sets in Baidu Netdisk.

Baidu network disk link:

Link: https://pan.baidu.com/s/1bAKoSG7VHE98JdHJPGJvcw

Extraction code: ibje

（1）NEU surface defect database

The data set collects 6 kinds of defects including inclusions, scratches, pressed scale, cracks, pits and plaques, with 300 images for each defect, and the image size is 200×200. The data set includes classification and target detection. However, there are a small number of errors in the labeling of target detection, which need attention. Some examples are as follows:

（2）Micro surface defect database

Miniature steel strip defect data, the defect is only about 6×6 pixels in size

（3）Oil pollution defect database

Dataset of Silicon Steel Surface Defects Interferenced by Oil Contamination

Strip defect data set provided by Severstal Steel Company in Kaggle

Four types of strip surface defects are provided in this dataset. There are 12568 training sets and 5506 testing sets. The image size is 1600×256.

Severstal Strip Defect Dataset

If there is a need for academic research, it can be used without restrictions, please designate PAO Severstal ( https://www.severstal.com/) as the dataset owner.

Official link: https://www.kaggle.com/c/severstal-steel-defect-detection/data

I originally wanted to upload it to Baidu Netdisk for everyone to download, but more than 500 files require super membership. . . Tens of thousands of pictures are too many to upload manually.

UCI Steel Plates Faults Data Set

The dataset contains 7 strip defect types. This data set is not image data, but 28 kinds of characteristic data of steel strip defects, which can be used in machine learning projects.

Official link: https://archive.ics.uci.edu/ml/datasets/Steel+Plates+Faults

Baidu network disk link:

Link: https://pan.baidu.com/s/1VhEW8xodv5XhTnBKoz5Z_w

Extraction code: 9uv2

DAGM 2007 dataset

The dataset is artificially generated and contains a total of 10 classes with an image size of 512×512. The data set is a partial example as follows:

Official link: https://hci.iwr.uni-heidelberg.de/content/weakly-supervised-learning-industrial-optical-inspection

Baidu network disk link:

Link: https://pan.baidu.com/s/1EyK3flXj2S9Uyooi10HsCA

Extraction code: j9qz

Magnetic Tile Defect Dataset

The data set collected by a research group of the Institute of Automation, Chinese Academy of Sciences is the data set of the paper "Saliency of magnetic tile surface defects". The images of 6 common magnetic tile defects were collected and marked with semantic segmentation.

Official link: https://github.com/abin24/Magnetic-tile-defect-datasets

Baidu network disk link:

Link: https://pan.baidu.com/s/1iuSHXoVJT-eInFh9xNzEyw

Extraction code: ky8i

Kolektor Surface Defects Dataset

This dataset is an electronic commutator defect dataset collected and labeled by Kolektor Group. The data set contains 50 written electronic commutators, each with 8 pictures and its semantically segmented label. The size of the image is 500×1240 pixels. For more convenient training, the image needs to be adjusted to 512×1408 in advance.

Official link: https://www.vicos.si/Downloads/KolektorSDD

Baidu network disk link:

Link: https://pan.baidu.com/s/1HSzHC1ltHvt1hSJh_IY4Jg

Extract code: 1zlb

Rail Surface Defects Dataset

The RSDDs dataset contains two types of datasets: the first is the Type I RSDDs dataset captured from the fast lane, which contains 67 challenging images. The second is a dataset of Type II RSDDs captured from ordinary/heavy transport tracks, which contains 128 challenging images.

Each image of the two datasets contains at least one defect, and the background is complex and noisy.

These defects in the RSDDs dataset have been flagged by some professional human observers in the field of orbital surface inspection.

Official link: http://icn.bjtu.edu.cn/Visint/resources/RSDDs.aspx

Baidu network disk link:

Link: https://pan.baidu.com/s/1svsnqL0r1kasVDNjppkEwg

Extraction code: nanr

Cement Road Cracks Dataset

It is mainly aimed at the crack detection of cement pavement, which can be used for classification, segmentation and detection

Official link: cuilimeng/CrackForest

Baidu network disk link:

Link: https://pan.baidu.com/s/19qEBt0JDLs1v6jS5y8HJhQ

Extraction code: 7nzx

Bridge Crack Image Data

Detection data of bridge cracks:

Baidu network disk link:

Link: https://pan.baidu.com/s/1z-y3GhsWmbqzezD-eZSL-A

Extraction code: z493

Concrete Surface Crack Defect Dataset

The dataset contains images of various concrete surfaces with and without cracks. The image data is divided into negative (no cracks) and positive (no cracks) parts in separate folders for image classification. There are 20,000 images per category, for a total of 40,000 images with RGB channels of 227 x 227 pixels.

Official link: https://www.kaggle.com/arunrk7/surface-crack-detection

Tianchi Aluminum Profile Surface Defect Dataset

The data set provided in the 2018 Ali Tianchi Competition, Guangdong Industrial Intelligent Manufacturing Big Data Innovation Competition-Intelligent Algorithm Competition.

Official link: [Flying Guangdong Cloud 2018] Guangdong Industrial Intelligent Manufacturing Big Data Innovation Competition-Intelligent Algorithm Competition Questions and Data-Tianchi Competition-Alibaba Cloud Tianchi

Baidu network disk link:

Link: https://pan.baidu.com/s/1jnSwJ2xRzdSplUtvTbuIuw

Extraction code: i10s

Tianchi Textile Surface Anomaly Dataset

In the actual production process of cloth, due to the influence of various factors, there will be stains, holes, hair particles and other defects. In order to ensure product quality, it is necessary to carry out defect detection on cloth. Cloth defect inspection is an important link in the production and quality management of the textile industry. At present, manual inspection is easily affected by subjective factors and lacks consistency; and the long-term work of inspectors under strong light has a great impact on vision. Due to the wide variety of cloth defects, various morphological changes, and the difficulty of observation and identification, intelligent detection of cloth defects has been a technical bottleneck that has plagued the industry for many years. This data covers various important defects of cloth in the textile industry, and each picture contains one or more defects. The data includes two types of plain cloth and patterned cloth. Among them, about 8,000 pieces of plain color cloth are used for the preliminary round; about 12,000 pieces of patterned cloth are used for the semi-finals.

Baidu network disk link:

Link: https://pan.baidu.com/s/1eRCCpQhkH05gBTaDkdSAAw

Extraction code: 2j46

AITEX Fabric Defect Dataset

This dataset contains 245 images of 4096x256 pixels with 7 different fabric structures. There are 140 defect-free images in the dataset, 20 of each type of fabric. In addition, there are 105 pictures of 12 different types of fabric defects commonly seen in the textile industry.

AITEX FABRIC IMAGE DATABASE - Aitexwww.aitex.es/afid/

Workpiece wear dataset: BSData

The dataset contains 1104 3-channel images and 394 image annotations, and the surface damage type is "pitting".

2Obe/BSData: Dataset for classification, detection and prognostics of surface defects on ball screw drives (github.com)github.com/2Obe/BSData

KTH-TIPS database

The background texture dataset provided by the Royal Institute of Technology in Sweden includes 10 categories including sandpaper, aluminum foil, styrofoam, sponge, corduroy, linen, cotton, black bread, orange peel and biscuits.

Baidu network disk link:

Link: https://pan.baidu.com/s/10essXdRrZtlx4CcSirq6Kw

Extraction code: am65

Printed Circuit Board (PCB) Defect Dataset

This is a public synthetic PCB dataset released by Peking University, which contains 1386 images with 6 types of defects (missing holes, mouse bites, opens, shorts, strays, pseudo-copper) for detection, classification and configuration. quasi-task.

Official link: Peking University Intelligent Robot Open Laboratory

Baidu network disk link: https://pan.baidu.com/s/1hoPNd7_SAxOWa2XbBZZuTg

Printed Circuit Board (PCB) Defect Dataset

DeepPCB: A data set contains 1500 pairs of images, each pair of images consists of a defect-free template image and an aligned test image, which annotated the 6 most common PCB defects: open, short, mousebite, spur, pin hole and spurious copper.

Surface-Defect-Detection/DeepPCB at master Charmve/Surface-Defect-Detectiongithub.com/Charmve/Surface-Defect-Detection/tree/master/DeepPCBUploading...ReuploadCancel

AITEX dataset

The database consists of 245 4096 x 256 pixel images of seven different fabric structures. There are 140 non-defect images in the database, 20 for each type of fabric, besides this, there are 105 images of different types of fabric defects (12 types of defects) that are common in the textile industry. The large size of the image allows the user to use different window sizes, thus increasing the number of samples. Databases on the Internet also contain segmentation masks of all images with defects such that white pixels represent defect regions and the remaining pixels are black.