Intrusion Detection Systems based on Artificial Intelligence need robust data sources in order to achieve strong generalization levels from the knowledge domain of interest. Anomaly detection is a well-known topic in cybersecurity, and its application to the Internet of Things can lead to suitable protection techniques against problems such as DoS and DDoS attacks.


Arbitrarily falling dices were photographed individually and monochromatically inside an Ulbricht sphere from two fixed perspectives. Overall, 11 dices with edge size 16 mm were used for 2133 falling experiments repeatedly. 5 of these dices were modified manually to have the following anomalies: drilled holes, missing dots, sawing gaps and scratches. All pictures in the uploaded pickle containers have a resolution of 400 times 400 pixels with normalized grey scale floating point values of 0 (black) through 1 (white).


The datasets contain files for training (“x_training.pickle”, w/o anomalies) and testing (“x_test.pickle”, w/ and w/o anomalies). Labels were saved in “y_test.pickle” whereas label zero correspond to non-anomalous data. Because the pose of the falling dice was not constrained the two fixed perspectives had the chance to see anomalies at all in 60 out of 100 experiments. Hence the test dataset contains 60 anomalous samples. Furthermore, data is augmented w.r.t. erased patches, changes in image constituents like brightness, and altered geometry like flipping and rotating.The shapes of the pickles are

  • w/o augmentation, x_train.pickle: (2000, 2, 400, 400)
  • w/o augmentation, x_test.pickle: (133, 2, 400, 400)
  • w/o augmentation, y_test.pickle: (133,)
  • w/ augmentation, x_train.pickle: (4000, 2, 400, 400)
  • w/ augmentation, x_test.pickle: (133, 2, 400, 400)
  • w/ augmentation, y_test.pickle: (133,)

Shoulder Physiotherapy Activity Recognition 9-Axis Dataset (SPARS9x) 

Suggested uses of this dataset include performing supervised classification analysis of physiotherapy exercises, or to perform out-of-distribution detection analysis with unlabeled activities of daily living data.

(Please note that this dataset is currently only available in individual csv files for each subject as we are having difficulties uploading zipped folders. We have reached out to IEEEDataport to help us resolve this issue)

This dataset consists of 20 csv files, one for each of the subjects in the dataset (e.g. "spars9x_s0.csv" corresponds to the first subject of the dataset). These are contained in the zip file

We have also provided the dataset in a numpy array format (, consisting of a single array and 20 subarrays, one for each dataset subject. These may be loaded into memory using the load() function of the python library numpy.

Either file needs to be decompressed (unzipped) before use.

Data Labels:

0: Unlabeled Activities of Daily Living (Non-Exercise)

1: External Rotation (Isometric)

2: External Rotation in 90 Degrees Abduction in Scapula Plane

3: Internal Rotation (Isometric)

4: Extension (Isometric)

5: Abduction (Isometric)

6: Biceps Muscle Strengthening

7: Cross Chest Adduction

8: Active Flexion

9: Shoulder Girdle Stabilization with Elevation

10: Triceps Pull Downs

Data Format:

Inertial data and heart rate have been interpolated to 50 Hz (inertial sensor acquisition was approximately 50 Hz, but sensors recorded asynchronously and so interpolation was needed).

Each record consists of an Nx12 array, where numbered columns correspond to:

0-2: Accelerometer (X/Y/Z) in G's

3-5: Magnetometer (X/Y/Z) in μT's

6-8: Gyroscope (X/Y/Z) in rad/s

9: Heart Rate in bpm

10: Shoulder (0 = Unlabeled, 1 = Left Shoulder, 2 = Right Shoulder)

11: Activity Label (0-10 as described above)

Note: All exercise data has left or right shoulder indicated, whereas other activity data does not. Other activity data may be assigned to left or right shoulder for analysis, or split between each as required.

Citation Request:

If using this data in academic publication, please cite the following manuscript:

Boyer, P.; Burns, D.; Whyne, C. Out-of-Distribution Detection of Human Activity Recognition with Smartwatch Inertial Sensors. Sensors 2021, 21, x.


This dataset is composed of 4-Dimensional time series files, representing the movements of all 38 participants during a novel control task. In the ‘’ file this can be set up to 6-Dimension, by the ‘fields_included’ variable. Two folders are included, one ready for preprocessing (‘subjects raw’) and the other already preprocessed ‘subjects preprocessed’.


This data set comprises 4223 videos from a laser surface heat treatment process (also called laser heat treatment) applied to cylindrical workpieces made of steel. The purpose of the dataset is to detect anomalies in the laser heat treatment learning a model from a set of non-anomalous videos.In the laser heat treatment, the laser beam is following a pattern similar to an "eight" with a frequency of 100 Hz. This pattern is sometimes modified to avoid obstacles in the workpieces.The videos are recorded at a frequency of 1000 frames per second with a thermal camera.


See for details on the structure of the dataset.