Sensing Anomalies as Potential Hazards: Datasets and Benchmarks

Mantegazza, Dario; Redondo, Carlos Anta; Espada, Fran; Gambardella, Luca Maria; Giusti, Alessandro; Guzzi, Jérôme

doi:10.1007/978-3-031-15908-4_17

Cited by 3 publications

(3 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We explored various hyperparameters of autoencoder, RNVP, BCLASS, WRN, and RNVP+OE. Most of the hyperparameters were set using a preliminary search or were inspired by previous research [1], [3], [5], [6]: the autoencoder's bottleneck size, learning rate, maximum number of epochs, input size and architecture details; the RNVP input size, learning rate, coupling layer size and number, and input masking; WRN's architecture and hyperparameters. For all experiments except AE and WRN+OE, we let the model run for 500 epochs, then choose the best performing model over the validation set.…”

Section: A Hyperparameter Explorationmentioning

confidence: 99%

See 1 more Smart Citation

An Outlier Exposure Approach to Improve Visual Anomaly Detection Performance for Mobile Robots

Mantegazza

Giusti

Gambardella

et al. 2022

IEEE Robot. Autom. Lett.

Self Cite

View full text Add to dashboard Cite

We consider the problem of building visual anomaly detection systems for mobile robots. Standard anomaly detection models are trained using large datasets composed only of nonanomalous data. However, in robotics applications, it is often the case that (potentially very few) examples of anomalies are available. We tackle the problem of exploiting these data to improve the performance of a Real-NVP anomaly detection model, by minimizing, jointly with the Real-NVP loss, an auxiliary outlier exposure margin loss. We perform quantitative experiments on a novel dataset (which we publish as supplementary material) designed for anomaly detection in an indoor patrolling scenario. On a disjoint test set, our approach outperforms alternatives and shows that exposing even a small number of anomalous frames yields significant performance improvements.

show abstract

Section: A Hyperparameter Explorationmentioning

confidence: 99%

“…The complete dataset, as well as a preliminary version presented in previous work[3], is available at https://github.com/idsia-robotics/ hazard-detection; it contains two additional scenarios of drones flying in indoor environments, which are not used in this paper.…”

mentioning

confidence: 99%

An Outlier Exposure Approach to Improve Visual Anomaly Detection Performance for Mobile Robots

Mantegazza

Giusti

Gambardella

et al. 2022

IEEE Robot. Autom. Lett.

Self Cite

View full text Add to dashboard Cite

show abstract

“…Considered anomalies include presence of humans, unexpected objects on the floor, defects to the robot. Preliminary versions of the dataset are used in [1 , 3] . This version is available at [12]…”

mentioning

confidence: 99%