Anomaly Classification

MVTec AD is a dataset for benchmarking anomaly detection methods with a focus on industrial inspection. It contains over 5000 high-resolution images divided into fifteen different object and texture categories. Each category comprises a set of defect-free training images and a test set of images with various kinds of defects as well as images without defects. There are two common metrics: Detection AUROC and Segmentation (or pixelwise) AUROC Detection (or, classification) methods output single float (anomaly score) per input test image. Segmentation methods output anomaly probability for each pixel. "To assess segmentation performance, we evaluate the relative per-region overlap of the segmentation with the ground truth. To get an additional performance measure that is independent of the determined threshold, we compute the area under the receiver operating characteristic curve (ROC AUC). We define the true positive rate as the percentage of pixels that were correctly classified as anomalous" [1] Later segmentation metric was improved to balance regions with small and large area, see PRO-AUC and other in [2] Source: MVTEC ANOMALY DETECTION DATASET Image Source: https://www.mvtec.com/company/research/datasets/mvtec-ad/ [1] Paul Bergmann et al, "MVTec AD — A Comprehensive Real-World Dataset for Unsupervised Anomaly Detection" [2] Bergmann, P., Batzner, K., Fauser, M. et al. The MVTec Anomaly Detection Dataset: A Comprehensive Real-World Dataset for Unsupervised Anomaly Detection. Int J Comput Vis (2021). https://doi.org/10.1007/s11263-020-01400-4

The VisA dataset contains 12 subsets corresponding to 12 different objects as shown in the above figure. There are 10,821 images with 9,621 normal and 1,200 anomalous samples. Four subsets are different types of printed circuit boards (PCB) with relatively complex structures containing transistors, capacitors, chips, etc. For the case of multiple instances in a view, we collect four subsets: Capsules, Candles, Macaroni1 and Macaroni2. Instances in Capsules and Macaroni2 largely differ in locations and poses. Moreover, we collect four subsets including Cashew, Chewing gum, Fryum and Pipe fryum, where objects are roughly aligned. The anomalous images contain various flaws, including surface defects such as scratches, dents, color spots or crack, and structural defects like misplacement or missing parts.

The GoodsAD dataset contains 6124 images with 6 categories of common supermarket goods. Each category contains multiple goods. All images are acquired with 3000 × 3000 high-resolution. The object locations in the images are not aligned. Most objects are in the center of the images and one image only contains a single object. Most anomalies occupy only a small fraction of image pixels. Both image-level and pixel-level annotations are provided. Each image is named with 6 digits, with the first three digits representing the category of the product and the last three representing the serial number. The dataset format is same as MVTec AD.

The dataset contains 2,624 samples of $300\times300$ pixels 8-bit grayscale images of functional and defective solar cells with varying degree of degradations extracted from 44 different solar modules. The defects in the annotated images are either of intrinsic or extrinsic type and are known to reduce the power efficiency of solar modules. All images are normalized with respect to size and perspective. Additionally, any distortion induced by the camera lens used to capture the EL images was eliminated prior to solar cell extraction.

Overview The edeniss2020 dataset is a time series dataset. It consists of equidistant sensor readings stemming from 97 sensors in the EDEN ISS research greenhouse. EDEN ISS was a (almost) closed loop research greenhouse build under the lead of the German Aerospace Center to study Controlled Environment Agriculture (CEA) techniques and plant growth for future long-term space missions. EDEN ISS was deployed in Antarctica in 2018 next to the german Neumayer III polar station and has been in operation for four years. The data contained in the edeniss2020 dataset was recorded during the third mission year in 2020 between 2020/01/01 and 2020/12/30. Every sensor within the dataset is related to one of the following Subsystems |Acronym | Description | |--|--| | AMS‑FEG | Atmosphere Management System (AMS) of the Future Exploration Greenhouse (FEG) | | AMS‑SES | Atmosphere Management System AMS of the Service Section (SES)| | ICS | Illumination Control System | | NDS | Nutrient Delivery System | | TCS | Thermal Control System| ## Specification |Item|Description | |--|--| | Number of Files | 97 | | Start Date | 2020/01/01 00:00:05 | | End Date | 2020/12/30 23:55:00 | | Sampling Rate | 1/300 Hz (5min) | ## Contents The dataset includes the following files: - ams-feg/*.csv: Sensor readings related to the AMS-FEG - ams-ses/*.csv: Sensor readings related to the AMS-SES - ics/*.csv: Temperature readings related to the ICS - nds/*.csv: Sensor readings related to the NDS - tcs/*.csv: Sensor readings related to the - edeniss2020.csv: Description and Units of the measurements. - README.md: This file. | Subsystem | Sensor | #sensors | Note | |--|--|--|--| | AMS-FEG | CO2 | 2 | | | | Photosynthetic Active Radiation (PAR) | 2 | | | | Relative Humidity (RH) | 2 | | | | Temperature (T) | 3 | | | AMS-SES | CO2 | 2 | | | Photosynthetic Active Radiation (PAR) | 1 | | | | Relative Humidity (RH) | 1 | | | | Temperature (T) | 3 | | | | Vapor Pressure Deficit (VPD) | 1 | | | ICS | Temperature (T) | 38 | Measured at the LED lamp above each growth tray | | | NDS | Electrical Conductivity (EC) | 4 | EC of the nutrient solutions | | | Level (H) | 2 | Level of the solution in the nutrient solution tanks | | | PH-Value (PH) | 4 | PH Value of the nutrient solutions | | | Pressure \(P\) | 8 | Pressure in the piping from the tanks to the growth racks in the FEG | | | Temperature (T) | 4 | Temperature of the nutrient solutions. | | | Volume (V) | 2 | Volume up to the level sensor | | TCS | Pressure \(P\) | 3 | | | Relative Humidity (RH) | 2 | | | Temperature (T) | 12 | | | Valve (VALVE) | 3 | ## Usage The dataset can be used for uni- and multivariate analysis. To read a uni-variate time series with pandas do: python import pandas as pd # Example of loading ams-feg/co2-1.csv data = pd.read_csv('./ams-feg/co2-1.csv') print(data.head()) To read the data for a whole subsystem (e.g. AMS-FEG) as a multivariate time series: python import pandas as pd import glob SUBSYSTEM = 'ams-feg' combined_df = pd.DataFrame() for i, file in enumerate(glob.glob(f'./{SUBSYSTEM}/*.csv')): df = pd.read_csv(file, header=0, usecols=[0, 1]) if i == 0: combined_df = df else: combined_df = pd.merge(combined_df, df, on='time') combined_df = combined_df.sort_values(by='time').reset_index(drop=True) print(combined_df) ## License Creative Commons Attribution 4.0 International (CC BY 4.0 Deed) ## Citation If you use this dataset, please cite it as follows: Rewicki, F., Norman, T., Vrakking, V., (2024). edeniss2020. Zenodo. http://doi.org/10.5281/zenodo.11485183 ## Contact Ferdinand Rewicki: ferdinand.rewicki@dlr.de ## Version History * v1.0 Initial Version

Hawk Annotation Dataset includes language descriptions specifically for anomaly scenes in seven existing video anomaly datasets. These seven datasets include a variety of anomalous scenarios such as crime (UCF-Cirme), campus (ShanghaiTech and CUHK Avenue), pedestrian walkways (UCSD Ped1 and Ped2), traffic (DoTA), and human behavior (UBnormal). With the support of these visual scenarios, this dataset can perform comprehensive fine-tuning for various abnormal scenarios, being closer to open-world scenarios.

MVTec-AC is a curated refinement of the widely-used MVTec-AD dataset, specifically designed for anomaly classification—distinguishing between different types of anomalies rather than merely detecting if an image is anomalous. While MVTec-AD focuses on binary detection and suffers from mislabeled or ambiguous samples, MVTec-AC introduces manually corrected labels and reorganized anomaly categories to enable robust multi-class evaluation. Key improvements include the correction of 36 misclassified samples, merging of 4 overlapping classes, removal of 4 ambiguous ‘combined’ classes, and exclusion of the toothbrush category, which contains only a single trivial anomaly type. These changes support consistent, fine-grained assessment of classification models in industrial visual inspection contexts.

VisA-AC is a refined benchmark based on the VisA dataset, tailored for the task of anomaly classification—distinguishing between different types of anomalies rather than simply detecting whether an image is anomalous. While the original VisA provides anomaly type information in an Excel file, it includes numerous under-sampled and ambiguous classes. VisA-AC addresses these issues by removing classes with fewer than 10 samples, merging visually similar categories, and manually correcting mislabeled samples. Additionally, anomaly classes in VisA-AC are organized into separate folders—following the structure of MVTec-AC—for easier integration and usage. The resulting dataset ensures both statistical robustness and semantic clarity, supporting rigorous evaluation of multi-class anomaly classification methods in real-world industrial settings.

The Lusitano dataset was collected over a 3-month period, spanning from January to March, from Paulo de Oliveira, S.A., a prominent textile company, based in Covilhã, Portugal, renowned for its innovative contributions to the textile industry. To collect the images for the dataset, we placed one camera in front of a fabric inspection machine, along with a strong and nearly uniform light source. This dataset comprises 4096 × 1024 images, captured by an industrial-grade Teledyne Dalsa Linea camera. The camera’s high resolution and precision ensure the accurate depiction of textile samples, with the level of detail necessary for defect analysis. None of the defects depicted in this dataset are artificially generated; they stem from genuine occurrences observed during this collection period, and thus represent real-world challenges encountered in textile production processes. The dataset also showcases normal images. We announce two folders, train and test in the same folder architecture, a MVTEC AD dataset. The training set contains 32k normal images, while the testing set contains 1646 defect images and 1038 normal images