Scanpath comparison metrics universally represent fixations as dimensionless (x, y) points, despite the human fovea subtending approximately 1° ∼ 2° of visual angle. This mismatch penalises spatially proximate fixations that process identical visual content. We propose FDISS (Foveal Disc IoU Scanpath Score), a metric that models each fixation as a foveal disc and computes similarity via bidirectional nearest-neighbour IoU matching, yielding interpretable Precision, Recall, and F1 scores. FDISS introduces a biologically grounded perceptual tolerance zone absent from all existing metrics and requires no free parameters at runtime.

• DemoFace provides a demographically balanced pixelated real face image dataset to mitigate biases in face biometric systems. • The dataset’s structured image–text embedding multimodality supports downstream tasks and facilitates the analysis of model biases in CVFMs. • Through our novel Responsible AI methodology, we developed the dataset and established a set of baselines using SOTA CVFMs to assess the performance and fairness of CVFMs in face biometric tasks. • Through our evaluation using both new and adapted metrics, we revealed inherent bias patterns in several SOTA CVFMs, providing valuable insights to guide future research. Bias and fairness are critical challenges in data-driven computer vision (CV), where limited demographic diversity in training data worsens these challenges. Face biometric (face recognition) systems are core tasks of CV that are highly impacted by these challenges, as existing real-face datasets lack comprehensive demographic representation, whereas current synthetic datasets promote stereotypes. CV Foundation Models (CVFMs) are currently at the forefront of CV applications, including face biometrics, which use global features in multimodal data. However, the scarcity of large-scale, demographic multimodal datasets, such as image-text embeddings for model fine-tuning (or training), limits the fairness in state-of-the-art (SOTA) CVFMs for downstream face biometric tasks. To address these issues, we introduce DemoFace, a balanced demographic face dataset comprising 30,240 pixelated real face images of 672 representative individuals evenly distributed across 48 demographic groups, categorized by ethnicity/race, gender, and age. We gathered images using an API set up from multiple copyright-free public forums. The collected images were then manually filtered, anonymized, and annotated by two independent research groups, and then lightly pixelated for privacy preservation. DemoFace’s image-text embedding multimodality enables fine-tuning (or training) of CVFMs for fairness-focused face biometrics tasks and bias pattern evaluation. Through two empirical studies: face authentication as classification and textual description as token generation, we established baseline scores across ethnicity/race, gender, and age groups. Our baselines identified inherent bias patterns through both new and tailored metrics derived from existing ones, emphasizing the need for more equitable AI models. Here is the Repository: Link

Anomaly detection in time series poses a critical challenge in industrial monitoring, environmental sensing, and infrastructure reliability, where accurately distinguishing anomalies from complex temporal patterns remains an open problem. While existing methods, such as the Anomaly Transformer leveraging multi-layer association discrepancy between prior and series distributions and Dual Attention Contrastive Representation Learning architecture (DCdetector) employing dual-attention contrastive learning, have advanced the field, critical limitations persist. These include sensitivity to short-term context windows, computational inefficiency, and degraded performance under noisy and non-stationary real-world conditions. To address these challenges, we present MAAT (Mamba Adaptive Anomaly Transformer), an enhanced architecture that refines association discrepancy modeling and reconstruction quality for more robust anomaly detection. Our work introduces two key contributions to the existing Anomaly transformer architecture: Sparse Attention, which computes association discrepancy more efficiently by selectively focusing on the most relevant time steps. This reduces computational redundancy while effectively capturing long-range dependencies critical for discerning subtle anomalies. A Mamba-Selective State Space Model (Mamba-SSM) is also integrated into the reconstruction module. A skip connection bridges the original reconstruction and the Mamba-SSM output, while a Gated Attention mechanism adaptively fuses features from both pathways. This design balances fidelity and contextual enhancement dynamically, improving anomaly localization and overall detection performance. Extensive experiments on benchmark datasets demonstrate that MAAT significantly outperforms prior methods, achieving superior anomaly distinguishability and generalization across diverse time series applications. By addressing the limitations of existing approaches, MAAT sets a new standard for unsupervised time series anomaly detection in real-world scenarios. Code available at https://github.com/ilyesbenaissa/MAAT . • MAAT uses Sparse Attention and Mamba-SSM for temporal dependencies. • Block-wise sparse attention reduces costs vs models like DCdetector . • Gated attention merges sparse attention with Mamba-SSM for better detection. • Uses SSM via Mamba for effective complex, noisy data management . • Outperforms Anomaly Transformer and DCdetector in F1, accuracy .

Evaluating house prices is crucial for various stakeholders, including homeowners, investors, and policymakers. However, traditional spatial interpolation methods have limitations in capturing the complex spatial relationships that affect property values. To address these challenges, we have developed a new method called Multi-Head Gated Attention for spatial interpolation. Our approach builds upon attention-based interpolation models and incorporates multiple attention heads and gating mechanisms to better capture spatial dependencies and contextual information. Importantly, our model produces embeddings that reduce the dimensionality of the data, enabling simpler models like linear regression to outperform complex ensembling models. We conducted extensive experiments to compare our model with baseline methods and the original attention based interpolation model. The results show a significant improvement in the accuracy of house price predictions, validating the effectiveness of our approach. This research advances the field of spatial interpolation and provides a robust tool for more precise house price evaluation. Our GitHub repository. 1 1 Final_file/tree/main/ASI-main . contains the data and code for all datasets, which are available for researchers and practitioners interested in replicating or building upon our work. • Multi-Head Gated Attention model improves spatial interpolation for house price prediction. • Geographical and structural attention mechanisms enhance spatial feature extraction. • Embedding generation reduces data complexity, boosting model performance and accuracy. • New dataset combines Italian city data with structural and geographical attributes. • Model outperforms traditional methods, validated across multiple diverse datasets.

The increasing use of deepfake technology enables the creation of realistic and deceptive content, raising concerns about several serious issues, including biometric authentication, misinformation, politics, privacy, and trust. Many Deepfake Detection (DD) models are entering the market to combat the misuse of deepfakes. With these developments, one primary issue occurs in ensuring the explainability of the proposed detection models to understand the rationale of the decision. This paper aims to investigate the state-of-the-art explainable DD models across multiple modalities, including image, video, audio, and text. Unlike existing surveys that focus on detection methodologies with minimal attention to explainability and limited modality coverage, this paper directly focuses on these gaps. It offers a comprehensive analysis of advanced explainability techniques, including Grad-CAM, LIME, SHAP, LRP, Saliency Maps, and Anchors, for detecting deceptive content across the modalities. It identifies the strengths and limitations of existing models and outlines research directions to enhance explainability and interpretability in future works. By exploring these models, we aim to enhance transparency, provide deeper insights into model decisions, and bridge the gap between detection accuracy with explainability in DD models.

Abstract In recent years, estimating beauty of faces has attracted growing interest in the fields of computer vision and machine learning. This is due to the emergence of face beauty datasets (such as SCUT-FBP, SCUT-FBP5500 and KDEF-PT) and the prevalence of deep learning methods in many tasks. The goal of this work is to leverage the advances in Deep Learning architectures to provide stable and accurate face beauty estimation from static face images. To this end, our proposed approach has three main contributions. To deal with the complicated high-level features associated with the FBP problem by using more than one pre-trained Convolutional Neural Network (CNN) model, we propose an architecture with two backbones (2B-IncRex). In addition to 2B-IncRex, we introduce a parabolic dynamic law to control the behavior of the robust loss parameters during training. These robust losses are ParamSmoothL1, Huber, and Tukey. As a third contribution, we propose an ensemble regression based on five regressors, namely Resnext-50, Inception-v3 and three regressors based on our proposed 2B-IncRex architecture. These models are trained with the following dynamic loss functions: Dynamic ParamSmoothL1, Dynamic Tukey, Dynamic ParamSmoothL1, Dynamic Huber, and Dynamic Tukey, respectively. To evaluate the performance of our approach, we used two datasets: SCUT-FBP5500 and KDEF-PT. The dataset SCUT-FBP5500 contains two evaluation scenarios provided by the database developers: 60-40% split and five-fold cross-validation. Our approach outperforms state-of-the-art methods on several metrics in both evaluation scenarios of SCUT-FBP5500. Moreover, experiments on the KDEF-PT dataset demonstrate the efficiency of our approach for estimating facial beauty using transfer learning, despite the presence of facial expressions and limited data. These comparisons highlight the effectiveness of the proposed solutions for FBP. They also show that the proposed Dynamic robust losses lead to more flexible and accurate estimators.

Since the emergence of the Covid-19 pandemic in late 2019, medical imaging has been widely used to analyze this disease. Indeed, CT-scans of the lungs can help diagnose, detect, and quantify Covid-19 infection. In this paper, we address the segmentation of Covid-19 infection from CT-scans. To improve the performance of the Att-Unet architecture and maximize the use of the Attention Gate, we propose the PAtt-Unet and DAtt-Unet architectures. PAtt-Unet aims to exploit the input pyramids to preserve the spatial awareness in all of the encoder layers. On the other hand, DAtt-Unet is designed to guide the segmentation of Covid-19 infection inside the lung lobes. We also propose to combine these two architectures into a single one, which we refer to as PDAtt-Unet. To overcome the blurry boundary pixels segmentation of Covid-19 infection, we propose a hybrid loss function. The proposed architectures were tested on four datasets with two evaluation scenarios (intra and cross datasets). Experimental results showed that both PAtt-Unet and DAtt-Unet improve the performance of Att-Unet in segmenting Covid-19 infections. Moreover, the combination architecture PDAtt-Unet led to further improvement. To Compare with other methods, three baseline segmentation architectures (Unet, Unet++, and Att-Unet) and three state-of-the-art architectures (InfNet, SCOATNet, and nCoVSegNet) were tested. The comparison showed the superiority of the proposed PDAtt-Unet trained with the proposed hybrid loss (PDEAtt-Unet) over all other methods. Moreover, PDEAtt-Unet is able to overcome various challenges in segmenting Covid-19 infections in four datasets and two evaluation scenarios.

The recognition of Covid-19 infection and distinguishing it from other Lung diseases from CT-scan is an emerging field in machine learning and computer vision community. In this paper, we proposed deep learning based approach to recognize the Covid-19 infection from the CT-scans. Our approach consists of two main stages. In the first stage, we trained deep learning architectures with Multi-task strategy for Slice-Level classification. In the second stage, we used the previous trained models with XG-boost classifier to classify the whole CT-scan into Normal, Covid-19 or Cap class. The evaluation of our approach achieved promising results on the validation data of SPGC-COVID dataset. In more details, our approach achieved 87.75% as overall accuracy and 96.36%, 52.63% and 95.83% sensitivities for Covid-19, Cap and Normal, respectively. From other hand, our approach achieved the fifth place on the three test datasets of SPGC on COVID-19 challenge where our approach achieved the best result for Covid-19 sensitivity.

Computer vision-based techniques are more and more employed in healthcare and medical fields nowadays, principally, as a support tool to medical staff in order to help them making quick and correct diagnosis. One of the hot topics in this arena concerns the automatic classification of skin lesions. Several promising works have been proposed in the last couple of years, mainly leveraging Convolutional Neural Networks (CNN). However, the proposed pipeline mainly rely on complex data pre-processing and there is no systematic investigation about how available deep models can actually reach the accuracy needed for real applications. In order to overcome these drawbacks, in this work, an end-to-end pipeline is introduced and some of the most recent Convolutional Neural Networks (CNNs) architectures are included in it and compared on the largest common benchmark dataset recently introduced. To this aim, for the first time in this application context, a new network design paradigm, namely RegNet, has been exploited to get the best models among a population of configurations. The paper introduces a threefold level of contribution and novelty with respect to the literature: the deep investigation of several CNN architectures driving to a consistent improvement of the lesions recognition accuracy, the exploitation of a new network design paradigm able to study the behavior of populations of models and a deep discussion about pros and cons of each analyzed method by paving the path towards new research lines.

Automatic recognition of products on grocery shelf images is a new and attractive topic in computer vision and machine learning since, it can be exploited in different application areas. This paper introduces a complete end-to-end pipeline, without preliminary radiometric and spatial transformations usually involved while dealing with the considered issue, and it provides a systematic investigation of recent machine learning models based on convolutional neural networks for addressing the product recognition task by exploiting the proposed pipeline on a recent challenging grocery product dataset. The investigated models were never been used in this context: they derive from the successful and more generic object recognition task and have been properly tuned to address this specific issue. Besides, also ensembles of nets built by most advanced theoretical fundaments have been taken into account. Gathered classification results were very encouraging since the recognition accuracy has been improved up to 15% with respect to the leading approaches in the state of art on the same dataset. A discussion about the pros and cons of the investigated solutions are discussed by paving the path towards new research lines.

Sea surveillance and monitoring is receiving an increasing attention in the last years especially in specific areas and countries, such as the Mediterranean sea. The automatic detection of boats is a strategic activity that can be realized by means of satellite or RGB images. Satellite monitoring requires expensive hardware (Lidar, Landscape, and so on) and high-level efforts and permissions; on the other hand, satellite images do not allow a high level of resolution able to provide information about the number of passengers on the boat, or its name/ID. The use of RGB cameras placed in specific coastal sites allows to realize a low-cost and scalable monitoring system. Convolutional Neural Networks (CNN) are commonly used in application of object detection, classification and tracking. In addition, they are used also for re-identification, mainly applied to people and cars, rarely to boats. The use of neural networks requires a specific training based on labeled datasets. At present there are just a few annotated datasets for boat detection, while nothing is present for boat re-identification. Here we propose a new dataset for boat re-identification; it is composed by annotated images of boats acquired during last summer in the Marine Protected Area (MPA) of Porto Cesareo. In addition, we propose as baseline the results obtained by the application of a Discriminatively Learned CNN for boat reidentification.

Iris segmentation is driven by three different quality factors: accuracy, usability and speed. Unfortunately the deeply analysis of the literature shows that the greatest efforts of the researchers mainly focus on accuracy and speed. Proposed solutions, in fact, do not meet the usability requirement since they are based on specific optimizations related to the operating context and they impose binding conditions on the sensors to be used for the acquisition of periocular images. This paper tries to fill this gap by introducing an innovative iris segmentation technique that can be used in unconstrained environments, under non-ideal imaging conditions and, above all, that does not require any interaction for adaptation to different operating conditions. Experimental results, carried out on challenging databases, demonstrate that the high usability of the proposed solution does not penalize segmentation accuracy which, in some respects, outperforms that of the leading approaches in the literature.

This paper presents a new method to automatically locate pupils in images (even with low-resolution) containing human faces. In particular pupils are localized by a two steps procedure: at first self-similarity information is extracted by considering the appearance variability of local regions and then they are combined with an estimator of circular shapes based on a modified version of the Circular Hough Transform. Experimental evidence of the effectiveness of the method was achieved on challenging databases containing facial images acquired under different lighting conditions and with different scales and poses.

The paper proposes a robust estimation method which implements, in cascade, two algorithms: (i) a Random Sample and Consensus (RANSAC) algorithm and (ii) a novel nonlinear prediction error estimator minimizing a cost function inspired by the mathematical definition of Gibbs entropy. The minimization of the nonlinear cost function allows to refine the Consensus Set found with standard RANSAC in order to reach optimal estimates of geometric transformation parameters under image stitching context. The method has been experimentally tested and compared with a standard RANSAC-MSAC algorithm where noticeable improvements are recorded in terms of computational complexity and quality of the stitching process, namely of the mean squared symmetric re-projection error.

This paper presents an inexpensive framework for 3-D seabed mosaic reconstruction, based on an asynchronous stereo vision system when simplifying motion assumptions are used. In order to achieve a metric reconstruction some knowledge about the scene is recovered by a simple and reliable calibration step. The major issue in calibration come from the asynchronism that complicate the proper frames selection. To overcome this problem a stereo frames selection based on epipolar gap evaluation (EGE) is proposed. Stereo disparity maps are evaluated by using both local and global approaches. To deal with brightness constancy model violation, zero-mean normalized cross-correlation is used as similarity measure in local approach, whereas a histogram equalization is necessary in global approach in order to improve min-cut based algorithms. Experimental results validate the proposed framework, allowing to define 3-D mosaics having visual quality similar to those obtained by using specialized hardware.

In this paper an automated video surveillance system for human posture recognition using active contours and neural networks is presented. Localization of moving objects in the scene and human posture estimation are key features of the proposed architecture. The system architecture consists of five sequential modules that include the moving target detection process, two levels of segmentation process for interested element localization, features extraction of the object shape and a human posture classification system based on the radial basis functions neural network. Moving objects are detected by using an adaptive background subtraction method with an automatic background adaptation speed parameter and a new fast gradient vector flow snake algorithm for the elements segmentation is proposed. The developed system has been tested for the classification of three different postures such as standing, bending and squatting considering different kinds of feature. Results are promising and the architecture is also useful for the discrimination of human activities.

Shadows detection is a fundamental step in visual surveillance and monitoring systems. Shadow points are often misclassified as object points causing errors in localization, segmentation, measurements, tracking and classification of moving objects. In order to detect potential shadow points an automatic segmentation procedure is performed and, for all moving pixels, the approach evaluates the compatibility of photometric properties with shadow characteristics. The approach is improved using a new method focused on the similarity between little textured patches: this method is based on the observation that shadow regions present same textural characteristics in each frame of the gray-level video-sequence and in the corresponding adaptive background model. We suggest a new approach to describe textural information in terms of redundant functions. The algorithm is designed to be unaffected by scene type, background type or light conditions. Results validate the algorithm 's performance on a benchmark suite of indoor and outdoor video sequences.

We present mechanisms for attention control and pattern categorization as the basis for robot cognition. For attention, we gather information from attentional feature maps extracted from sensory data constructing salience maps to decide where to foveate. For identification, multi-feature maps are used as input to an associative memory, allowing the system to classify a pattern representing a region of interest. As a practical result, our robotic platforms are able to select regions of interest and perform shifts of attention focusing on the selected regions, and to construct and maintain attentional maps of the environment in an efficient manner.

In this paper a solution to a manipulation control problem (target identification and grasping) using a real platform in combination with a vision system is proposed. The task for the end-effector is to approach a randomly placed spherical object of known size. The control law is approximated by using an approach based on fuzzy logic. The controller determines the parameters of the target (a spherical object acquired from the vision system mounted on the manipulator) using a vision algorithm. The fuzzy rules are built in a supervised approach after studying the behavior of the system. Experimental results obtained using an industrial manipulator (PUMA 560 with 6-DOF) are presented.

We describe a navigation approach, based on a reinforcement learning algorithm, that allows a mobile robot to move in an unknown indoor environment learning autonomously in a few trials the actions for reaching a particular goal location. The control architecture merges visual information and sonar readings to evaluate the state of the system to which the learning algorithm relates the best movement for reaching the goal. As a result, after limited experience, the robot learns efficient behavioral sequences and improves its learning incrementally. In addition the learning system adapts well to new situations of the environment or new task requirements. The results obtained in simulation and in real experiments, carried out in our laboratory, have shown that our system is tolerant to noise in sensor measurements and during each trial is always able to reach the goal.