Deepfake detectors remain vulnerable to transfer-based black-box attacks, in which adversarial examples are generated on a source surrogate model and transferred to a target model, unknown to the attacker. Yet how source–target compatibility shapes attack success remains poorly understood. Prior studies evaluate limited detector pools and rarely disentangle architectural from training factors. We conduct a controlled evaluation of adversarial transferability across 60 detectors spanning six backbones, two pretraining regimes, and five training-data configurations, using two attack procedures: AutoAttack (AA) and the Carlini–Wagner attack with Expectation over Transformation (CW–EOT). Matched comparisons reveal significantly higher transfer when source and target share an exact backbone, architecture family, pretraining regime, or training data. This compatibility structure is attack-dependent: exact backbone compatibility has the largest effect under AA, whereas shared pretraining and training data have the largest effects under CW–EOT. When transfer is averaged across non-target sources, mean attack success rate (ASR) is 7.21% under AA and 19.52% under CW–EOT. By contrast, a multi-source oracle combining both attacks attains a 64.48% mean ASR after excluding exact backbone and training-data matches, showing that source averaging can substantially understate target vulnerability. We release 240,000 adversarially perturbed images, complete pairwise transfer results, detector configurations, and evaluation code. These findings establish source–target compatibility and source-model selection as central dimensions of credible transfer-based black-box robustness evaluation.

Rapid Eye Movement (REM) sleep is marked by intense brain activity coupled with muscular atonia. When this mechanism fails, abnormal behaviors may occur, often indicating REM Sleep Behavior Disorder (RBD) and serving as an early marker of neurodegenerative diseases. Reliable confirmation of such events requires both polysomnographic (PSG) signals and video observation, but synchronizing these modalities outside laboratory settings remains a challenge. This work presents a MATLAB application that integrates European Data Format (EDF) signals with MP4 recordings through an intuitive graphical interface. The system enables simultaneous navigation of electrophysiological data and video, supported by signal preprocessing, artifact reduction, and timeline synchronization. Researchers can use the tool to align multimodal recordings and collaboratively review events with clinicians, ensuring more consistent interpretation. By bridging technical and clinical perspectives, the application reduces manual workload, supports longitudinal studies, and promotes reproducibility in multimodal sleep research.

This study investigates the use of SHAP (SHapley Additive exPlanations) values as an explainable artificial intelligence (xAI) technique applied on a facial attribute classification task. We analyse the consistency of SHAP value distributions across diverse classifier architectures that share the same feature extractor, revealing that key features driving attribute classification remain stable regardless of classifier architecture. Our findings highlight the challenges in interpreting SHAP values at the individual sample level, as their reliability depends on the model’s ability to learn distinct class-specific features; models exploiting inter-class correlations yield less representative SHAP explanations. Furthermore, pixel-level SHAP analysis reveals that superior classification accuracy does not necessarily equate to meaningful semantic understanding; notably, despite FaceNet exhibiting lower performance than CLIP, it demonstrated a more nuanced grasp of the underlying class attributes. Finally, we address the computational scalability of SHAP, demonstrating that KernelExplainer becomes infeasible for high-dimensional pixel data, whereas DeepExplainer and GradientExplainer offer more practical alternatives with trade-offs. Our results suggest that SHAP is most effective for small to medium feature sets, providing interpretable and computationally manageable explanations.

The rapid advancement of wireless technologies, particularly Wi-Fi, has spurred significant research into indoor human activity detection across various domains (e.g., healthcare, security, and industry). This work explores the non-invasive and cost-effective Wi-Fi paradigm and the application of deep learning for human activity recognition using Wi-Fi signals. Focusing on the challenges in machine interpretability, motivated by the increase in data availability and computational power, this paper uses explainable artificial intelligence to understand the inner workings of transformer-based deep neural networks designed to estimate human pose (i.e., human skeleton key points) from Wi-Fi channel state information. Using different strategies to assess the most relevant sub-carriers (i.e., rollout attention and masking attention) for the model predictions, we evaluate the performance of the model when it uses a given number of sub-carriers as input, selected randomly or by ascending (high-attention) or descending (low-attention) order. We concluded that the models trained with fewer (but relevant) sub-carriers are competitive with the baseline (trained with all sub-carriers) but better in terms of computational efficiency (i.e., processing more data per second).

This study investigates the effects of occlusions on the fairness of face recognition systems, particularly focusing on demographic biases. Using the Racial Faces in the Wild (RFW) dataset and synthetically added realistic occlusions, we evaluate their effect on the performance of face recognition models trained on the BUPT-Balanced and BUPT-GlobalFace datasets. We note increases in the dispersion of FMR, FNMR, and accuracy alongside decreases in fairness according to Equalized Odds, Demographic Parity, STD of Accuracy, and Fairness Discrepancy Rate. Additionally, we utilize a pixel attribution method to understand the importance of occlusions in model predictions, proposing a new metric, Face Occlusion Impact Ratio (FOIR), that quantifies the extent to which occlusions affect model performance across different demographic groups. Our results indicate that occlusions exacerbate existing demographic biases, with models placing higher importance on occlusions in an unequal fashion across demographics.

Bias has been a constant in face recognition models. Over the years, researchers have looked at it from both the model and the data point of view. However, their approach to mitigation of data bias was limited and lacked insight on the real nature of the problem. Here, in this document, we propose to revise our use of ethnicity labels as a continuous variable instead of a discrete value per identity. We validate our formulation both experimentally and theoretically, showcasing that not all identities from one ethnicity contribute equally to the balance of the dataset; thus, having the same number of identities per ethnicity does not represent a balanced dataset. We further show that models trained on datasets balanced in the continuous space consistently outperform models trained on data balanced in the discrete space. We trained more than 65 different models, and created more than 20 subsets of the original datasets.

Leveraging the capabilities of Knowledge Distillation (KD) strategies, we devise a strategy to fight the recent retraction of face recognition datasets. Given a pretrained Teacher model trained on a real dataset, we show that carefully utilising synthetic datasets, or a mix between real and synthetic datasets to distil knowledge from this teacher to smaller students can yield surprising results. In this sense, we trained 33 different models with and without KD, on different datasets, with different architectures and losses. And our findings are consistent, using KD leads to performance gains across all ethnicities and decreased bias. In addition, it helps to mitigate the performance gap between real and synthetic datasets. This approach addresses the limitations of synthetic data training, improving both the accuracy and fairness of face recognition models.

The development of deep learning algorithms has extensively empowered humanity's task automatization capacity. However, the huge improvement in the performance of these models is highly correlated with their increasing level of complexity, limiting their usefulness in human-oriented applications, which are usually deployed in resource-constrained devices. This led to the development of compression techniques that drastically reduce the computational and memory costs of deep learning models without significant performance degradation. These compressed models are especially essential when implementing multi-model fusion solutions where multiple models are required to operate simultaneously. This paper aims to systematize the current literature on this topic by presenting a comprehensive survey of model compression techniques in biometrics applications, namely quantization, knowledge distillation and pruning. We conduct a critical analysis of the comparative value of these techniques, focusing on their advantages and disadvantages and presenting suggestions for future work directions that can potentially improve the current methods. Additionally, we discuss and analyze the link between model bias and model compression, highlighting the need to direct compression research toward model fairness in future works.

As in school, one teacher to cover all subjects is insufficient to distill equally robust information to a student. Hence, each subject is taught by a highly specialised teacher. Following a similar philosophy, we propose a multiple specialized teacher framework to distill knowledge to a student network. In our approach, directed at face recognition use cases, we train four teachers on one specific ethnicity, leading to four highly specialized and biased teachers. Our strategy learns a project of these four teachers into a common space and distill that information to a student network. Our results highlighted increased performance and reduced bias for all our experiments. In addition, we further show that having biased/specialized teachers is crucial by showing that our approach achieves better results than when knowledge is distilled from four teachers trained on balanced datasets. Our approach represents a step forward to the understanding of the importance of ethnicity-specific features.

Synthetic data is gaining increasing popularity for face recognition technologies, mainly due to the privacy concerns and challenges associated with obtaining real data, including diverse scenarios, quality, and demographic groups, among others. It also offers some advantages over real data, such as the large amount of data that can be generated or the ability to customize it to adapt to specific problem-solving needs. To effectively use such data, face recognition models should also be specifically designed to exploit synthetic data to its fullest potential. In order to promote the proposal of novel Generative AI methods and synthetic data, and investigate the application of synthetic data to better train face recognition systems, we introduce the 2nd FRCSyn-onGoing challenge, based on the 2nd Face Recognition Challenge in the Era of Synthetic Data (FRCSyn), originally launched at CVPR 2024. This is an ongoing challenge that provides researchers with an accessible platform to benchmark i) the proposal of novel Generative AI methods and synthetic data, and ii) novel face recognition systems that are specifically proposed to take advantage of synthetic data. We focus on exploring the use of synthetic data both individually and in combination with real data to solve current challenges in face recognition such as demographic bias, domain adaptation, and performance constraints in demanding situations, such as age disparities between training and testing, changes in the pose, or occlusions. Very interesting findings are obtained in this second edition, including a direct comparison with the first one, in which synthetic databases were restricted to DCFace and GANDiffFace.

Face recognition technology, despite its advancements and increasing accuracy, still presents significant challenges in explainability and ethical concerns, especially when applied in sensitive domains such as surveillance, law enforcement, and access control. The opaque nature of deep learning models jeopardises transparency, bias, and user trust. Concurrently, the proliferation of web applications presents a unique opportunity to develop accessible and interactive tools for demonstrating and analysing these complex systems. These tools can facilitate model decision exploration with various images, aiding in bias mitigation or enhancing users' trust by allowing them to see the model in action and understand its reasoning. We propose an explainable face recognition web application designed to support enrolment, identification, authentication, and verification while providing visual explanations through pixel-wise importance maps to clarify the model's decision-making process. The system is built in compliance with the European Union General Data Protection Regulation, ensuring data privacy and user control over personal information. The application is also designed for scalability, capable of efficiently managing large datasets. Load tests conducted on databases containing up to 1,000,000 images confirm its efficiency. This scalability ensures robust performance and a seamless user experience even with database growth.

Face recognition (FR) systems are vulnerable to morphing attacks, which refer to face images created by morphing the facial features of two different identities into one face image to create an image that can match both identities, allowing serious security breaches. In this work, we apply a frequency-based explanation method from the area of explainable face recognition to shine a light on how FR models behave when processing a bona fide or attack pair from a frequency perspective. In extensive experiments, we used two different state-of-the-art FR models and six different morphing attacks to investigate possible differences in behavior. Our results show that FR models rely differently on different frequency bands when making decisions for bona fide pairs and morphing attacks. In the following step, we show that this behavioral difference can be used to detect morphing attacks in an unsupervised setup solely based on the observed frequency-importance differences in a generalizable manner.

Synthetic data is gaining increasing relevance for training machine learning models. This is mainly motivated due to several factors such as the lack of real data and intra-class variability, time and errors produced in manual labeling, and in some cases privacy concerns, among others. This paper presents an overview of the 2nd edition of the Face Recognition Challenge in the Era of Synthetic Data (FRCSyn) organized at CVPR 2024. FRCSyn aims to investigate the use of synthetic data in face recognition to address current technological limitations, including data privacy concerns, demographic biases, generalization to novel scenarios, and performance constraints in challenging situations such as aging, pose variations, and occlusions. Unlike the 1st edition, in which synthetic data from DCFace and GANDiffFace methods was only allowed to train face recognition systems, in this 2nd edition we propose new sub-tasks that allow participants to explore novel face generative methods. The outcomes of the 2nd FRCSyn Challenge, along with the proposed experimental protocol and benchmarking contribute significantly to the application of synthetic data to face recognition.

Face recognition applications have grown in parallel with the size of datasets, complexity of deep learning models and computational power. However, while deep learning models evolve to become more capable and computational power keeps increasing, the datasets available are being retracted and removed from public access. Privacy and ethical concerns are relevant topics within these domains. Through generative artificial intelligence, researchers have put efforts into the development of completely synthetic datasets that can be used to train face recognition systems. Nonetheless, the recent advances have not been sufficient to achieve performance comparable to the state-of-the-art models trained on real data. To study the drift between the performance of models trained on real and synthetic datasets, we leverage a massive attribute classifier (MAC) to create annotations for four datasets: two real and two synthetic. From these annotations, we conduct studies on the distribution of each attribute within all four datasets. Additionally, we further inspect the differences between real and synthetic datasets on the attribute set. When comparing through the Kullback-Leibler divergence we have found differences between real and synthetic samples. Interestingly enough, we have verified that while real samples suffice to explain the synthetic distribution, the opposite could not be further from being true.

The proliferation of explicit material online, particularly pornography, has emerged as a paramount concern in our society. While state-of-the-art pornography detection models already show some promising results, their decision-making processes are often opaque, raising ethical issues. This study focuses on uncovering the decision-making process of such models, specifically fine-tuned convolutional neural networks and transformer architectures. We compare various explainability techniques to illuminate the limitations, potential improvements, and ethical implications of using these algorithms. Results show that models trained on diverse and dynamic datasets tend to have more robustness and generalisability when compared to models trained on static datasets. Additionally, transformer models demonstrate superior performance and generalisation compared to convolutional ones. Furthermore, we implemented a privacy-preserving framework during explanation retrieval, which contributes to developing secure and ethically sound biometric applications.

Morphing attacks keep threatening biometric systems, especially face recognition systems. Over time they have become simpler to perform and more realistic, as such, the usage of deep learning systems to detect these attacks has grown. At the same time, there is a constant concern regarding the lack of interpretability of deep learning models. Balancing performance and interpretability has been a difficult task for scientists. However, by leveraging domain information and proving some constraints, we have been able to develop IDistill, an interpretable method with state-of-the-art performance that provides information on both the identity separation on morph samples and their contribution to the final prediction. The domain information is learnt by an autoencoder and distilled to a classifier system in order to teach it to separate identity information. When compared to other methods in the literature it outperforms them in three out of five databases and is competitive in the remaining.

Morphing attacks keep threatening biometric systems, especially face recognition systems. Over time they have become simpler to perform and more realistic, as such, the usage of deep learning systems to detect these attacks has grown. At the same time, there is a constant concern regarding the lack of interpretability of deep learning models. Balancing performance and interpretability has been a difficult task for scientists. However, by leveraging domain information and proving some constraints, we have been able to develop IDistill, an interpretable method with state-of-the-art performance that provides information on both the identity separation on morph samples and their contribution to the final prediction. The domain information is learnt by an autoencoder and distilled to a classifier system in order to teach it to separate identity information. When compared to other methods in the literature it outperforms them in three out of five databases and is competitive in the remaining.

In the context of biometrics, matching confidence refers to the confidence that a given matching decision is correct. Since many biometric systems operate in critical decision-making processes, such as in forensics investigations, accurately and reliably stating the matching confidence becomes of high importance. Previous works on biometric confidence estimation can well differentiate between high and low confidence, but lack interpretability. Therefore, they do not provide accurate probabilistic estimates of the correctness of a decision. In this work, we propose a probabilistic interpretable comparison (PIC) score that accurately reflects the probability that the score originates from samples of the same identity. We prove that the proposed approach provides optimal matching confidence. Contrary to other approaches, it can also optimally combine multiple samples in a joint PIC score which further increases the recognition and confidence estimation performance. In the experiments, the proposed PIC approach is compared against all biometric confidence estimation methods available on four publicly available databases and five state-of-the-art face recognition systems. The results demonstrate that PIC has a significantly more accurate probabilistic interpretation than similar approaches and is highly effective for multi-biometric recognition. The code is publicly-available 1 .

With the ever-growing complexity of deep learning models for face recognition, it becomes hard to deploy these systems in real life. Researchers have two options: 1) use smaller models; 2) compress their current models. Since the usage of smaller models might lead to concerning biases, compression gains relevance. However, compressing might be also responsible for an increase in the bias of the final model. We investigate the overall performance, the performance on each ethnicity subgroup and the racial bias of a State-of-the-Art quantization approach when used with synthetic and real data. This analysis provides a few more details on potential benefits of performing quantization with synthetic data, for instance, the reduction of biases on the majority of test scenarios. We tested five distinct architectures and three different training datasets. The models were evaluated on a fourth dataset which was collected to infer and compare the performance of face recognition models on different ethnicity.

VISIGRAPP 2021 proceedings on computer vision, imaging, and computer graphics theory and applications, deep learning for visual, etc.