Patch Memory Bank k-NN for Semi-supervised Visual Hazard Detection in Indoor Mobile Robots

Patch Memory Bank k-NN for Semi-supervised Visual Hazard Detection in Indoor Mobile Robots

————
**License**
————
Dataset is available under the CC BY 4.0 license https://creativecommons.org/licenses/by/4.0/.

————
**Summary**
————
The project provides a semi-supervised anomaly detection framework for indoor mobile robots using global and patch-level features extracted from CNN and Vision Transformer backbones. Hazards are treated as visual anomalies compared to a memory bank of patch features collected from normal data.

It is obligatory to cite the following paper in every work that uses the dataset:
Wozniak, P., Krzeszowski, T. (2026). Patch Memory Bank K-NN for Semi-supervised Visual Hazard Detection in Indoor Mobile Robots. In: Paszynski, M., Barnard, A.S., Zhang, Y.J. (eds) Computational Science – ICCS 2026 Workshops. ICCS 2026. Lecture Notes in Computer Science, vol 16789. Springer, Cham. https://doi.org/10.1007/978-3-032-29918-5_26

————
**Data description**
————
The dataset was specifically prepared for evaluation purposes, consisting of scenes recorded by a real mobile robot and including natural variability due to environmental changes. To create controlled evaluation scenarios, the dataset was extended with synthetically augmented data. Our approach integrates real-world and synthetically augmented data to enable systematic and robust assessment of~anomaly detection performance for mobile robots under diverse conditions. The dataset was created based on The Multi-Domain Dataset for Robots (MDDRobots) (https://doi.org/10.1038/s41597-025-05124-3).

————
**Dataset Structure**
————
– DataSet_RobotPiCamera_RGB_train – training data from the MDDRobots dataset (https://doi.org/10.1038/s41597-025-05124-3)
– DataSet_RobotPiCamera_RGB_hazards – test data with synthetically generated hazards
– Code – implementation of the method described in the paper
 
————
**Further information**
————
For any questions, comments or other issues please contact Piotr Woźniak <p.wozniak@prz.edu.pl>.