Deep Convolutional Neural Networks can be fooled with imperceptible image perturbations. This negatively affects the security of intelligent systems. It is especially important for applications, in which security is crucial, such as Automated Guided Vehicles (AGVs) and production systems. A great majority of adversarial attacks target the classification layer of the network. These attacks focus only on changing the prediction, but it is also beneficial to corrupt the deep data representations. It can result in higher confusion of neural networks, which is important for more in-depth security testing. Classifier-independent attacks could be used in the future to test other types of networks frequently used in industry, production and AGVs, e.g. object detectors. In this paper, we present Network Saturation Attack (NetSat) - a novel classifier-independent, white-box, non-targeted adversarial attack. It aims to ‘saturate’ the final convolutional feature maps with introduced patterns and thus - make the network unable to provide reasonable predictions. Classifier-independence results in samples with predictions further away from the true label and high flexibility of the method. To assess its harmfulness, we propose Dissimilarity Metric (DM). It describes how far is the predicted label from the true one in the class space. The proposed methods, due to their simplicity, flexibility and practicality, are suitable for testing of real-life systems. We release a GitHub repository with code and examples: https://github.com/iitis/NetSat-NetworkSaturationAttack.

Popularity of mobile vision applications constantly increases. These Artificial Intelligence technology-driven solutions exhibit high potential for the dynamically-changing landscape of education, in which they could be used as powerful tools to engage and motivate learners. The integration of game-like elements into educational software can be a very effective approach to capture the attention of students by providing interactive and immersive learning experiences. That is why in this study we present a category recognition method for computationally-limited devices such as smartphones to be used in a cost-efficient manner in the areas connected to interactive education. We use small and efficient ImageNet-trained Convolutional Neural Networks and Deep Template Matching to create a classification solution. To test our solution, we construct a specialized image dataset with different toys and items that could be suitable for early-stage education to engage young learners. We perform an extensive study on how the number of images used to create templates and different similarity/distance measures applied for matching impact the accuracy to provide practical observations that can be used to create real-life game-like elements for educational software. Our results indicate that utilizing more templates and similarity measures results in better generalization, representativeness, and robustness, particularly when it comes to rotation, which is crucial for mobile applications. We release our data and code via our GitHub repository at https://github.com/iitis/EduToyz.

Constantly evolving landscape of modern education makes the integration of technology crucial to enable more interactive and immersive learning experiences. One of the key problems in this domain is Human Activity Recognition (HAR), which uses standard smartphone sensors to understand and recognize user movements. While HAR exhibits high potential in this area, its persistent challenge is to differentiating between similar activities, whose occurrences can often lead to misclassification and reduced quality of applications. To tackle this problem, we employ specialized classifiers and use them in a hierarchical manner instead of using monolith multi-class models. We propose HierAct - a Hierarchical model that solves the multi-class human activity recognition problem. We also release a new HAR dataset - EduAct - (https://github.com/iitis/HierAct-Dataset) focused on activities that could be used to create game-like educational applications.Our results show that the performance of the proposed model is substantially better than other state-of-the-art machine learning models. Our work shows the potential of hierarchical classifiers in HAR applications for education, by offering educators and students a more accurate, reliable, and engaging interactive experience.

Classification is the key task of deep learning. Among others, it is used in computer vision for object recognition, and in natural language processing for the masked language modeling. All of these tasks are crucial for modern automatic and semi-automatic production facilities powered by Automated Guided Vehicles (AGVs), because they enable intelligent inspection, maintenance log analysis and operation control. Especially in such safety-critical domains, in which humans and machines often coexist, it is crucial to provide new methods that could help us understand the operation of these black-boxes. That is why we present a novel visual-analytic methods that can be used to examine how networks perceive relations/similarities between the known classes. Our methods operate solely on the trained models and do not require any data samples. They can also reveal some quality issues in the training datasets and indicate low model accuracy. Our methods are suitable for generic vision and language models, but can also be used in transfer learning scenarios. To empirically validate our approach in such a scenario, we conduct experiments on a state-of-the-art mobile vision model - MobileNetV2 - fine-tuned for vehicle classification. We release a new dataset - UtilityVehicles - featuring images of various vehicles that can occur in industry. The presented use case is a vision-based application for Augmented Reality applications for smartphones and embedded devices for AGVs in automatic and semi-automatic production facilities. We release a GitHub repository with data and code: https://github.com/iitis/UtilityVehicles.

Convolutional Neural Networks (CNNs) enabled breakthroughs in computer vision. They are also used in different domains of smart cities to process and analyze large amounts of image data, which is crucial for intelligent decision-making. CNNs trained on large-scale datasets such as ImageNet, possess remarkable image classification abilities. As it takes a lot of time and computational resources to train models on such datasets, we propose a solution to reuse these models in the area of smart-cities. We call our approach the model recycling, as it uses the existing versatile knowledge of ImageNet-trained networks for resource conservation and faster deployment of applications in the smart-city domains. For that purpose, we utilize the semantic connections between ImageNet categories to determine the sets of classes that can be used to create specialized classification models for smart transportation, shopping and education with no additional training. We present a methodology to extract such specialized models from generic CNNs trained on ImageNet. Such models can be used at low budget to create solutions for automatic data annotation and in interactive applications. They can also be used as a starting point for fine-tuning. We also present 2 strategies of creating ensembles with these specialized models for better overall and per-class accuracy. Our general methodology can be also used in other domains outside the smart cities.

Automated Ground Vehicles (AGVs) use deep-learning-based vision systems to perceive the surrounding environment and extract relevant information about it. Although deep learning models offer high capabilities, they require large amounts of data to be properly trained and tested. Testing is especially important when off-the-shelf models are used by the AGVs - to examine whether they can meet the demands of complex environments such as the production halls of automated factories. One area of such perception algorithms is object recognition. To test such systems, we propose a solution based on ArUco fiducial markers used for automatic labeling of objects. Our solution can be used to test deep learning systems in real time directly on a robot. Our solution requires minimal interference with the environment and additional infrastructure - the desired objects only need to be marked with a marker printed on a home printer. Therefore, the presented testing procedure can be used for testing of AGVs in real-life environments during a real ride from an actual robot perspective. Data gathered during the online testing can be used for the offline comparison of the accuracy of different deep learning models. Although we focus on the online and offline testing in our study, we also incorporated a marker masking procedure. Therefore, the resulting datasets may also be used for training.