Diffusion models conditioned on identity embeddings enable the generation of synthetic face images that consistently preserve identity across multiple samples. Recent work has shown that introducing an additional negative condition through classifier-free guidance during sampling provides a mechanism to suppress undesired attributes, thus improving inter-class separability. Building on this insight, we propose a dynamic weighting scheme for the negative condition that adapts throughout the sampling trajectory. This strategy leverages the complementary strengths of positive and negative conditions at different stages of generation, leading to more diverse yet identity-consistent synthetic data.

Face Recognition (FR) is used in a variety of application domains, from entertainment and; to security and surveillance. Such applications rely on the FR model to be robust and perform well in a variety of settings. To achieve this, state-of-the-art FR models typically use expressive adaptive margin loss functions, which tie the feature norm to concepts related to sample quality, such as recognizability and perceptual image quality. Recently, through the development of Face Image Quality Assessment (FIQA) techniques, biometric utility has become the preferred measure of face-image quality and has been shown to be a better predictor of the usefulness of samples for face recognition compared to more human-centric aspects, such as resolution, blur, and lighting, tied to general image quality. While image quality expressed through feature norms exhibits a certain level of correlation with biometric utility, it does not fully encapsulate all aspects of utility. To address this point, we propose a new adaptive margin loss, FunFace (Face Recognition Through Utility and Norm Estimation), which incorporates biometric utility, estimated by the Certainty Ratio, into the adaptive margin, taking inspiration from AdaFace. We show that FunFace (when used to train a face recognition model) achieves competitive results to other state-of-the-art FR models on benchmarks containing high-quality samples, while surpassing them on low quality benchmarks. The code is available at https://github.com/LSIbabnikz/FunFace.

This study investigates the impact of face image background correction through segmentation on face recognition and morphing attack detection performance in realistic, unconstrained image capture scenarios. The motivation is driven by operational biometric systems such as the European Entry/Exit System (EES), which require facial enrolment at airports and other border crossing points where controlled backgrounds usually required for such captures cannot always be guaranteed, as well as by accessibility needs that may necessitate image capture outside traditional office environments. By analyzing how such preprocessing steps influence both recognition accuracy and security mechanisms, this work addresses a critical gap between usability-driven image normalization and the reliability requirements of large-scale biometric identification systems. Our study evaluates a comprehensive range of segmentation techniques, three families of morphing attack detection methods, and four distinct face recognition models, using databases that include both controlled and in-the-wild image captures. The results reveal consistent patterns linking segmentation to both recognition performance and face image quality. Additionally, segmentation is shown to systematically influence morphing attack detection performance. These findings highlight the need for careful consideration when deploying such preprocessing techniques in operational biometric systems. https://github.com/EduardaCaldeira/FSB-BR.

Recent advances in Vision Transformers (ViTs) for face recognition (FR) have moved beyond the standard CLS-token paradigm. In this paradigm, a special classification token (CLS) is prepended to the patch embeddings and used as a representation of the input for downstream tasks. An alternative approach, Concatenated Patch Embeddings (CPE), instead leverages all patch tokens by concatenating them into a single vector, which is then projected into a compact face representation. CPE has been shown to improve recognition performance in comparison to CLS-based ones, but our qualitative analysis of attention maps showed the presence of artifacts that limit their interpretability. To address this issue, we incorporate register tokens, learnable tokens concatenated to the initial patch embeddings, and processed jointly through the ViT encoder blocks. This mechanism has been shown to produce more structured and interpretable attention maps compared to baseline ViT. We empirically demonstrate that these artifacts consistently appear across various ViT backbones, including small and large models, and that introducing register tokens effectively mitigates them. Adding four or eight registers significantly enhances interpretability, with eight registers providing the highest verification accuracies and smoothest attention structures. Our resulting model, ViT-8R, corresponds to a CPE-based ViT-B architecture augmented with eight register tokens achieves state-of-the-art performance among ViT-based FR models on large-scale IJB-B and IJBC benchmarks. Also, ViT-8R produces substantially clearer attention maps compared with the baseline model, which offer deeper insight into the model's attention behavior (https://github.com/TaharChettaoui/ViT-FR-Registers).

Vision Transformers (ViTs) have gained significant attention in computer vision and shown strong potential for face recognition (FR). However, their high computational cost makes deployment on resource-constrained devices challenging, motivating the need for methods that balance efficiency and accuracy. In this work, we investigate early exiting in pretrained ViTs as a simple yet effective training-free strategy for efficient FR inference. Leveraging the uniform feature dimensionality across transformer encoder blocks, we introduce ViT-FREE, a multi-exit framework that enables face verification directly from intermediate representations without modifying or retraining the backbone model, and thus, reducing inference cost. Our approach is motivated by the intrinsic behavior of ViTs, where transformer blocks iteratively refine representations within a shared representation space. Empirically, we show that patch embeddings and attention maps evolve progressively across depth, exhibiting high similarity between consecutive ViT blocks and increasing alignment with the final representation. This indicates gradual feature refinement and attention convergence, suggesting that intermediate layers already provide stable and discriminative representations suitable for early exiting. Through extensive experiments on multiple FR benchmarks, we systematically analyze the accuracy-efficiency trade-off across exit depths. Our results demonstrate that later exits achieve a highly favorable balance, with exiting at layer 10 yielding up to a 20% speedup while incurring only a ∼1.5% drop in verification performance on benchmarks such as IJB-C. Also, we propose ViT-FREEFT, a lightweight exit-specific finetuning strategy that adapts only the projection layers using a small synthetic dataset while keeping the transformer backbone frozen. This approach improves the performance of shallow exits while preserving the efficiency benefits and leaving deeper exits largely unaffected. Code and pretrained models will be publicly released.

This paper introduces SEEKr, a novel approach for improving the efficiency of Knowledge Distillation (KD) in Face Recognition (FR) models. State-of-the-art FR systems achieve strong performance. However, they incur substantial computational costs, particularly during the transfer of knowledge from large, high-capacity teacher models (T) to compact student models (S) using massive training datasets containing millions of samples. SEEKr addresses this challenge by accelerating KD through the selection of a compact yet highly informative subset of the training data. The central insight of SEEKr is to prioritize hard samples, inputs that are challenging for the student, during the distillation process. Emphasizing these samples allows the student to more accurately approximate the teacher's decision function while learning from diverse and informative examples. In contrast, easy samples are shown to contribute marginally to effective knowledge transfer and can be safely excluded without notable performance degradation. By selectively distilling knowledge from hard samples, SEEKr significantly reduces the number of training samples required for KD, thus substantially lowering computational cost. Extensive experiments on multiple large-scale Face Recognition datasets, across diverse teacher architectures and sample selection strategies, demonstrate that SEEKr consistently achieves competitive or superior distillation performance while significantly reducing computational overhead. The proposed method is general, easily integrable into existing KD frameworks, and offers a scalable solution for efficient Face Recognition model training in resource-constrained environments. Code and pretrained models are available at https://github.com/jankolf/SEEKr.

Face Image Quality Assessment (FIQA) is crucial for reliable face recognition (FR) systems, yet existing Vision Transformer (ViT)-based approaches rely exclusively on final-layer representations, ignoring quality-relevant information captured at intermediate network depths. This paper presents the first comprehensive investigation of how intermediate representations within ViTs contribute to face quality assessment through early exit mechanisms and score fusion strategies. We systematically analyze all twelve transformer blocks of ViT-FIQA architectures, demonstrating that different depths capture distinct and complementary quality-relevant information, as evidenced by varying attention patterns and performance characteristics across network layers. Leveraging these insights, we propose a score fusion framework that combines quality predictions from multiple transformer blocks without architectural modifications or additional training. Our early exit analysis reveals optimal performance-efficiency tradeoffs, enabling significant computational savings while maintaining competitive performance. Through extensive evaluation across eight benchmark datasets using four FR models, we demonstrate that our fusion strategy improves upon single-exit approaches. Our proposed quality fusion approach employs depth-weighted averaging that assigns progressively higher importance to deeper transformer blocks, achieving the best quality assessment performance by effectively leveraging the hierarchical nature of feature learning in ViTs. Our work challenges the conventional wisdom that only deep features matter for face analysis, revealing that intermediate representations contain valuable information for quality assessment. The proposed framework offers practical benefits for realworld biometric systems by enabling adaptive computation based on resource constraints while maintaining competitive quality assessment capabilities. The implementation is publicly available at: https://github.com/gurayozgur/EX-FIQA.

Face Image Quality Assessment (FIQA) aims to assess the recognition utility of face samples and is essential for reliable face recognition (FR) systems. Existing approaches require computationally expensive procedures such as multiple forward passes, backpropagation, or additional training, and only recent work has focused on the use of Vision Transformers. Recent studies highlighted that these architectures inherently function as saliency learners with attention patterns naturally encoding spatial importance. This work proposes ATTN-FIQA, a novel training-free approach that investigates whether presoftmax attention scores from pre-trained Vision Transformerbased face recognition models can serve as quality indicators. We hypothesize that attention magnitudes intrinsically encode quality: high-quality images with discriminative facial features enable strong query-key alignments producing focused, highmagnitude attention patterns, while degraded images generate diffuse, low-magnitude patterns. ATTN-FIQA extracts presoftmax attention matrices from the final transformer block, aggregate multi-head attention information across all patches, and compute image-level quality scores through simple averaging, requiring only a single forward pass through pre-trained models without architectural modifications, backpropagation, or additional training. Through comprehensive evaluation across eight benchmark datasets and four FR models, this work demonstrates that attention-based quality scores effectively correlate with face image quality and provide spatial interpretability, revealing which facial regions contribute most to quality determination. The implementation is publicly available at: https://github.com/gurayozgur/ATTN-FIQA.

Face recognition models have made substantial progress due to advances in deep learning and the availability of large-scale datasets. However, reliance on massive annotated datasets introduces challenges related to training computational cost and data storage, as well as potential privacy concerns regarding managing large face datasets. This paper presents DiffProb, the first data pruning approach for the application of face recognition. DiffProb assesses the prediction probabilities of training samples within each identity and prunes the ones with identical or close prediction probability values, as they are likely reinforcing the same decision boundaries, and thus contribute minimally with new information. We further enhance this process with an auxiliary cleaning mechanism to eliminate mislabeled and label-flipped samples, boosting data quality with minimal loss. Extensive experiments on CASIA-WebFace with different pruning ratios and multiple benchmarks, including LFW, CFP-FP, and IJB-C, demonstrate that DiffProb can prune up to $50 \%$ of the dataset while maintaining or even, in some settings, improving the verification accuracies. Additionally, we demonstrate DiffProb’s robustness across different architectures and loss functions. Our method significantly reduces training cost and data volume, enabling efficient face recognition training and reducing the reliance on massive datasets and their demanding management. The code, pretrained models, and pruned datasets are publicly released: https://github.com/EduardaCaldeira/DiffProb.

The major improvements in face recognition (FR) in recent years have been supported by large face databases. However, there are concerns about the legal and ethical aspects of using large authentic databases with the proper consent from individuals being questioned. Motivated by this, and by the technical need for larger and more diverse data, synthetic datasets are being increasingly used, taking advantage of recent advances in the field of generative models. A major challenge there is ensuring the generation of synthetic face images with realistic and controllable class separability. In this paper, we aim to enhance class separability, which is commonly low in GAN-based synthetic FR data and affects synthetic based FR performance. To achieve that, we propose a novel label smoothing scheme within a class-conditional generation process. The smoothing aims at going beyond hard labels that induce a class label to the generation, by pushing the generation process away from other classes. In extensive experiments, we show the benefit of label smoothing in the generative setup by showing increased class separability. This is also reflected in the models trained on the proposed data by outperforming its hard label baseline and the state-of-the-art GAN-based synthetic-based FR approaches on multiple established verification benchmarks.

Recent advances in generative modeling have enabled the generation of high-quality synthetic data that is applicable in a variety of domains, including face recognition. Here, state-of-the-art generative models typically rely on conditioning and fine-tuning of powerful pretrained diffusion models to facilitate the synthesis of realistic images of a desired identity. Yet, these models often do not consider the identity of subjects during training, leading to poor consistency between generated and intended identities. In contrast, methods that employ identity-based training objectives tend to overfit on various aspects of the identity, and in turn, lower the diversity of images that can be generated. To address these issues, we present in this paper a novel generative diffusion-based framework, called ID-Booth. ID-Booth consists of a denoising network responsible for data generation, a variational auto-encoder for mapping images to and from a lower-dimensional latent space and a text encoder that allows for prompt-based control over the generation procedure. The framework utilizes a novel triplet identity training objective and enables identity-consistent image generation while retaining the synthesis capabilities of pretrained diffusion models. Experiments with a state-of-the-art latent diffusion model and diverse prompts reveal that our method facilitates better intra-identity consistency and inter-identity separability than competing methods, while achieving higher image diversity. In turn, the produced data allows for effective augmentation of small-scale datasets and training of betterperforming recognition models in a privacy-preserving manner. The source code for the ID-Booth framework is publicly available at https://github.com/dariant/ID-Booth.

Foundation models are predominantly trained in an unsupervised or self-supervised manner on highly diverse and large-scale datasets, making them broadly applicable to various downstream tasks. In this work, we investigate for the first time whether such models are suitable for the specific domain of face recognition (FR). We further propose and demonstrate the adaptation of these models for FR across different levels of data availability, including synthetic data. Extensive experiments are conducted on multiple foundation models and datasets of varying scales for training and fine-tuning, with evaluation on a wide range of benchmarks. Our results indicate that, despite their versatility, pre-trained foundation models tend to underperform in FR in comparison with similar architectures trained specifically for this task. However, fine-tuning foundation models yields promising results, often surpassing models trained from scratch, particularly when training data is limited. For example, after fine-tuning only on 1K identities, DINOv2 ViT-S achieved average verification accuracy on LFW, CALFW, CPLFW, CFP-FP, and AgeDB30 benchmarks of 87.10%, compared to 64.70% achieved by the same model and without fine-tuning. While training the same model architecture, ViT-S, from scratch on 1k identities reached 69.96%. With access to larger-scale FR training datasets, these performances reach 96.03% and 95.59% for the DINOv2 and CLIP ViT-L models, respectively. In comparison to the ViT-based architectures trained from scratch for FR, fine-tuned same architectures of foundation models achieve similar performance while requiring lower training computational costs and not relying on the assumption of extensive data availability. We further demonstrated the use of synthetic face data, showing improved performances over both pre-trained foundation and ViT models. Additionally, we examine demographic biases, noting slightly higher biases in certain settings when using foundation models compared to models trained from scratch. We release our code and pre-trained models’ weights at github.com/TaharChettaoui/FRoundation . • First work to investigate the performance of foundation models for face recognition. • First to adapt foundation models using low-rank adaptation (LoRA) for face recognition. • Investigate the impact of dataset size on ViTs and foundation models tuned with LoRA for face recognition. • Evaluate foundation models and adapted ones on mainstream face recognition benchmarks. • Explore the potential of synthetic face datasets for fine-tuning foundation models and training ViTs.

Recent advances in deep face recognition have spurred a growing demand for large, diverse, and manually annotated face datasets. Acquiring authentic, high-quality data for face recognition has proven to be a challenge, primarily due to privacy concerns. Large face datasets are primarily sourced from web-based images, lacking explicit user consent. In this paper, we examine whether and how synthetic face data can be used to train effective face recognition models with reduced reliance on authentic images, and thus mitigating large authentic data collection concerns. First, we explored the performance gap among recent state-of-the-art face recognition models, trained only on synthetic data or authentic data. Then, we deepened our analysis by training a state-of-the-art back-bone with various combinations of synthetic and authentic data, gaining insights into optimizing the limited use of the latter for verification accuracy. Finally, we assessed the effectiveness of data augmentation approaches on synthetic and authentic data, with the same goal in mind. Our results highlighted the effectiveness of FR trained on combined datasets, particularly when combined with appropriate augmentation techniques.

Large-scale face recognition datasets are collected by crawling the Internet and without individuals' consent, raising legal, ethical, and privacy concerns. With the recent advances in generative models, recently several works proposed generating synthetic face recognition datasets to mitigate concerns in web-crawled face recognition datasets. This paper presents the summary of the Synthetic Data for Face Recognition (SDFR) Competition held in conjunction with the 18th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2024) and established to investigate the use of synthetic data for training face recognition models. The SDFR competition was split into two tasks, allowing participants to train face recognition systems using new synthetic datasets and/or existing ones. In the first task, the face recognition backbone was fixed and the dataset size was limited, while the second task provided almost complete freedom on the model backbone, the dataset, and the training pipeline. The submitted models were trained on existing and also new synthetic datasets and used clever methods to improve training with synthetic data. The submissions were evaluated and ranked on a diverse set of seven benchmarking datasets. The paper gives an overview of the submitted face recognition models and reports achieved performance compared to baseline models trained on real and synthetic datasets. Furthermore, the evaluation of submissions is extended to bias assessment across different demography groups. Lastly, an outlook on the current state of the research in training face recognition models using synthetic data is presented, and existing problems as well as potential future directions are also discussed.

Synthetic data is emerging as a substitute for authentic data to solve ethical and legal challenges in handling authentic face data. The current models can create real-looking face images of people who do not exist. However, it is a known and sensitive problem that face recognition systems are susceptible to bias, i.e. performance differences between different demographic and non-demographics attributes, which can lead to unfair decisions. In this work, we investigate how the diversity of synthetic face recognition datasets compares to authentic datasets, and how the distribution of the training data of the generative models affects the distribution of the synthetic data. To do this, we looked at the distribution of gender, ethnicity, age, and head position. Furthermore, we investigated the concrete bias of three recent synthetic-based face recognition models on the studied attributes in comparison to a baseline model trained on authentic data. Our results show that the generator generate a similar distribution as the used training data in terms of the different attributes. With regard to bias, it can be seen that the synthetic-based models share a similar bias behavior with the authentic-based models. However, with the uncovered lower intra-identity attribute consistency seems to be beneficial in reducing bias.

Current periocular and face recognition approaches utilize computationally costly deep neural networks, achieving notable recognition accuracies. Deploying such solutions in applications with limited computational resources requires minimizing their computational demand while maintaining similar recognition accuracies. Model compression techniques like model quantization can be used to reduce the computational costs of deep models. This approach is widely studied and applied to different machine-learning tasks, however it is understudied and investigated for biometrics. We propose in this work to reduce the computational cost of face and periocular recognition models using fixed- and mixed-precision model quantization. Specifically, we first quantize the full-precision models to fixed 8 and 6 bits, reducing the required memory footprint by 5x while maintaining, to a very large degree, the recognition accuracies. However, our achieved results demonstrated that by quantizing the models to extremely low b bits, e.g., below 6 bits, the accuracies significantly dropped, which motivated our investigation on mixed-precision quantization. Hence, we propose to utilize an iterative mixed-precision quantization scheme. In each iteration, the least important parameters are selected based on their weight magnitude and quantized to low b-bit precision and the model is fine-tuned. This approach is repeated until all parameters are quantized to low b-bit precision, achieving extreme reduction in memory footprint, e.g., 16x times, without significant loss in the model accuracies. The effectiveness of mixed- and fixed-precision quantization for biometric recognition models is studied and proved for two modalities, face and periocular, using three different deep network architectures and using different b bit precision.

The performance of state-of-the-art face recognition systems depends crucially on the availability of large-scale training datasets. However, increasing privacy concerns nowadays accompany the collection and distribution of biometric data, which has already resulted in the retraction of valuable face recognition datasets. The use of synthetic data represents a potential solution, however, the generation of privacy-preserving facial images useful for training recognition models is still an open problem. Generative methods also remain bound to the visible spectrum, despite the benefits that multispectral data can provide. To address these issues, we present a novel identity-conditioned generative framework capable of producing large-scale recognition datasets of visible and near-infrared privacy-preserving face images. The framework relies on a novel identity-conditioned dual-branch style-based generative adversarial network to enable the synthesis of aligned high-quality samples of identities determined by features of a pretrained recognition model. In addition, the framework incorporates a novel filter to prevent samples of privacy-breaching identities from reaching the generated datasets and improve both identity separability and intra-identity diversity. Extensive experiments on six publicly available datasets reveal that our framework achieves competitive synthesis capabilities while preserving the privacy of real-world subjects. The synthesized datasets also facilitate training more powerful recognition models than datasets generated by competing methods or even small-scale real-world datasets. Employing both visible and near-infrared data for training also results in higher recognition accuracy on real-world visible spectrum benchmarks. Therefore, training with multispectral data could potentially improve existing recognition systems that utilize only the visible spectrum, without the need for additional sensors.

Abstract Recent advancements in ubiquitous computing have emphasized the need for privacy-preserving occupancy detection in smart environments to enhance security. This work presents a novel occupancy detection solution utilizing privacy-aware sensing technologies. The solution analyzes time-series data to detect not only occupancy as a binary problem, but also determines whether one or multiple individuals are present in an indoor environment. On three real-world datasets, our models outperformed various state-of-the-art algorithms, achieving F1-scores up to 94.91% in single-occupancy detection and a macro F1-score of 91.55% in multi-occupancy detection. This makes our approach a promising solution for improving security in smart environments.

Face image quality assessment (FIQA) estimates the utility of the captured image in achieving reliable and accurate recognition performance. This work proposes a novel FIQA method, CR-FIQA, that estimates the face image quality of a sample by learning to predict its relative classifiability. This classifiability is measured based on the allocation of the training sample feature representation in angular space with respect to its class center and the nearest negative class center. We experimentally illustrate the correlation between the face image quality and the sample relative classifiability. As such property is only observable for the training dataset, we propose to learn this property by probing internal network observations during the training process and utilizing it to predict the quality of unseen samples. Through extensive evaluation experiments on eight benchmarks and four face recognition models, we demonstrate the superiority of our proposed CR-FIQA over state-of-the-art (SOTA) FIQA algorithms.11https://github.com/fdbtrs/CR-FIQA

Over the past years, the main research innovations in face recognition focused on training deep neural networks on large-scale identity-labeled datasets using variations of multi-class classification losses. However, many of these datasets are retreated by their creators due to increased privacy and ethical concerns. Very recently, privacy-friendly synthetic data has been proposed as an alternative to privacy-sensitive authentic data to comply with privacy regulations and to ensure the continuity of face recognition research. In this paper, we propose an unsupervised face recognition model based on unlabeled synthetic data (USynthFace). Our proposed USynthFace learns to maximize the similarity between two augmented images of the same synthetic instance. We enable this by a large set of geometric and color transformations in addition to GAN-based augmentation that contributes to the USynthFace model training. We also conduct numerous empirical studies on different components of our USynthFace. With the proposed set of augmentation operations, we proved the effectiveness of our USynthFace in achieving relatively high recognition accuracies using unlabeled synthetic data. The training code and pretrained model are publicly available under https://github.com/fdbtrs/Unsupervised-Face-Recognition-using-Unlabeled-Synthetic-Data.

Deep generative models have recently presented impressive results in generating realistic face images of random synthetic identities. To generate multiple samples of a certain synthetic identity, previous works proposed to disentangle the latent space of GANs by incorporating additional supervision or regularization, enabling the manipulation of certain attributes. Others proposed to disentangle specific factors in unconditional pretrained GANs latent spaces to control their output, which also requires supervision by attribute classifiers. Moreover, these attributes are entangled in GAN’s latent space, making it difficult to manipulate them without affecting the identity information. We propose in this work a framework, ExFaceGAN, to disentangle identity information in pretrained GANs latent spaces, enabling the generation of multiple samples of any synthetic identity. Given a reference latent code of any synthetic image and latent space of pretrained GAN, our ExFaceGAN learns an identity directional boundary that disentangles the latent space into two sub-spaces, with latent codes of samples that are either identity similar or dissimilar to a reference image. By sampling from each side of the boundary, our ExFaceGAN can generate multiple samples of synthetic identity without the need for designing a dedicated architecture or supervision from attribute classifiers. We demonstrate the generalizability and effectiveness of ExFaceGAN by integrating it into learned latent spaces of three SOTA GAN approaches. As an example of the practical benefit of our ExFaceGAN, we empirically prove that data generated by ExFaceGAN can be successfully used to train face recognition models (https://github.com/fdbtrs/ExFaceGAN).

This paper presents the summary of the Sclera Segmentation and Joint Recognition Benchmarking Competition (SSRBC 2023) held in conjunction with IEEE International Joint Conference on Biometrics (IJCB 2023). Different from the previous editions of the competition, SSRBC 2023 not only explored the performance of the latest and most advanced sclera segmentation models, but also studied the impact of segmentation quality on recognition performance. Five groups took part in SSRBC 2023 and submitted a total of six segmentation models and one recognition technique for scoring. The submitted solutions included a wide variety of conceptually diverse deep-learning models and were rigorously tested on three publicly available datasets, i.e., MASD, SBVPI and MOBIUS. Most of the segmentation models achieved encouraging segmentation and recognition performance. Most importantly, we observed that better segmentation results always translate into better verification performance.

The paper provides a summary of the 2023 Unconstrained Ear Recognition Challenge (UERC), a benchmarking effort focused on ear recognition from images acquired in uncontrolled environments. The objective of the challenge was to evaluate the effectiveness of current ear recognition techniques on a challenging ear dataset while analyzing the techniques from two distinct aspects, i.e., verification performance and bias with respect to specific demographic factors, i.e., gender and ethnicity. Seven research groups participated in the challenge and submitted a seven distinct recognition approaches that ranged from descriptor-based methods and deep-learning models to ensemble techniques that relied on multiple data representations to maximize performance and minimize bias. A comprehensive investigation into the performance of the submitted models is presented, as well as an in-depth analysis of bias and associated performance differentials due to differences in gender and ethnicity. The results of the challenge suggest that a wide variety of models (e.g., transformers, convolutional neural networks, ensemble models) is capable of achieving competitive recognition results, but also that all of the models still exhibit considerable performance differentials with respect to both gender and ethnicity. To promote further development of unbiased and effective ear recognition models, the starter kit of UERC 2023 together with the baseline model, and training and test data is made available from: http://ears.fri.uni-lj.si/

This paper presents the summary of the Efficient Face Recognition Competition (EFaR) held at the 2023 International Joint Conference on Biometrics (IJCB 2023). The competition received 17 submissions from 6 different teams. To drive further development of efficient face recognition models, the submitted solutions are ranked based on a weighted score of the achieved verification accuracies on a diverse set of benchmarks, as well as the deployability given by the number of floating-point operations and model size. The evaluation of submissions is extended to bias, cross-quality, and large-scale recognition benchmarks. Overall, the paper gives an overview of the achieved performance values of the submitted solutions as well as a diverse set of baselines. The submitted solutions use small, efficient network architectures to reduce the computational cost, some solutions apply model quantization. An outlook on possible techniques that are underrepresented in current solutions is given as well.

State-of-the-art face recognition (FR) systems are based on overparameterized deep neural networks (DNN) which commonly use face images with 2563colors. The use of DNN and the storage of face images as references for comparison are limited in resource-restricted domains, which are hemmed in storage and computational capacity. A possible solution is to store the image only as a feature, which renders the human evaluation of the image impossible and forces the use of a single DNN (vendor) across systems. In this paper, we present a novel study on the possibility and effect of image color quantization on FR performance and storage efficiency. We leverage our conclusions to propose harmonizing the color quantization with the low-bit quantization of FR models. This combination significantly reduces the bits required to represent both the image and the FR model. In an extensive experiment on diverse sets of DNN architectures and color quantization steps, we validate on multiple benchmarks that the proposed methodology can successfully reduce the number of bits required for image pixels and DNN data while maintaining nearly equal recognition rates. The code and pre-trained models are available at https://github.com/jankolf/ColorQuantization.

The availability of large-scale authentic face databases has been crucial to the significant advances made in face recognition research over the past decade. However, legal and ethical concerns led to the recent retraction of many of these databases by their creators, raising questions about the continuity of future face recognition research without one of its key resources. Synthetic datasets have emerged as a promising alternative to privacy-sensitive authentic data for face recognition development. However, recent synthetic datasets that are used to train face recognition models suffer either from limitations in intra-class diversity or cross-class (identity) discrimination, leading to less optimal accuracies, far away from the accuracies achieved by models trained on authentic data. This paper targets this issue by proposing IDiff-Face, a novel approach based on conditional latent diffusion models for synthetic identity generation with realistic identity variations for face recognition training. Through extensive evaluations, our proposed synthetic-based face recognition approach pushed the limits of state-of-the-art performances, achieving, for example, 98.00% accuracy on the Labeled Faces in the Wild (LFW) benchmark, far ahead from the recent synthetic-based face recognition solutions with 95.40% and bridging the gap to authentic-based face recognition with 99.82% accuracy*.

Bias and fairness of biometric algorithms have been key topics of research in recent years, mainly due to the societal, legal and ethical implications of potentially unfair decisions made by automated decision-making models. A considerable amount of work has been done on this topic across different biometric modalities, aiming at better understanding the main sources of algorithmic bias or devising mitigation measures. In this work, we contribute to these efforts and present the first study investigating bias and fairness of sclera segmentation models. Although sclera segmentation techniques represent a key component of sclera-based biometric systems with a considerable impact on the overall recognition performance, the presence of different types of biases in sclera segmentation methods is still underexplored. To address this limitation, we describe the results of a group evaluation effort (involving seven research groups), organized to explore the performance of recent sclera segmentation models within a common experimental framework and study performance differences (and bias), originating from various demographic as well as environmental factors. Using five diverse datasets, we analyze seven independently developed sclera segmentation models in different experimental configurations. The results of our experiments suggest that there are significant differences in the overall segmentation performance across the seven models and that among the considered factors, ethnicity appears to be the biggest cause of bias. Additionally, we observe that training with representative and balanced data does not necessarily lead to less biased results. Finally, we find that in general there appears to be a negative correlation between the amount of bias observed (due to eye color, ethnicity and acquisition device) and the overall segmentation performance, suggesting that advances in the field of semantic segmentation may also help with mitigating bias.

Recent deep face recognition models proposed in the literature utilized large-scale public datasets such as MS-Celeb-1M and VGGFace2 for training very deep neural networks, achieving state-of-the-art performance on mainstream benchmarks. Recently, many of these datasets, e.g., MS-Celeb-1M and VGGFace2, are retracted due to credible privacy and ethical concerns. This motivates this work to propose and investigate the feasibility of using a privacy-friendly synthetically generated face dataset to train face recognition models. Towards this end, we utilize a class-conditional generative adversarial network to generate class-labeled synthetic face images, namely SFace. To address the privacy aspect of using such data to train a face recognition model, we provide extensive evaluation experiments on the identity relation between the synthetic dataset and the original authentic dataset used to train the generative model. Our reported evaluation proved that associating an identity of the authentic dataset to one with the same class label in the synthetic dataset is hardly possible. We also propose to train face recognition on our privacy-friendly dataset, SFace, using three different learning strategies, multi-class classification, label-free knowledge transfer, and combined learning of multi-class classification and knowledge transfer. The reported evaluation results on five authentic face benchmarks demonstrated that the privacy-friendly synthetic dataset has a high potential to be used for training face recognition models, achieving, for example, a verification accuracy of 91.87% on LFW using multi-class classification and 99.13% using the combined learning strategy. The training code and the synthetic face image dataset are publicly released11https://github.com/fdbtrs/SFace-Privacy-friendly-and-Accurate-Face-Recognition-using-Synthetic-Data.

The supervised-learning-based morphing attack detection (MAD) solutions achieve outstanding success in dealing with attacks from known morphing techniques and known data sources. However, given variations in the morphing attacks, the performance of supervised MAD solutions drops significantly due to the insufficient diversity and quantity of the existing MAD datasets. To address this concern, we propose a completely unsupervised MAD solution via self-paced anomaly detection (SPL-MAD) by leveraging the existing large-scale face recognition (FR) datasets and the unsupervised nature of convolutional autoencoders. Using general FR datasets that might contain unintentionally and unlabeled manipulated samples to train an autoencoder can lead to a diverse reconstruction behavior of attack and bona fide samples. We analyze this behavior empirically to provide a solid theoretical ground for designing our unsupervised MAD solution. This also results in proposing to integrate our adapted modified self-paced learning paradigm to enhance the reconstruction error separability between the bona fide and attack samples in a completely unsupervised manner. Our experimental results on a diverse set of MAD evaluation datasets show that the proposed unsupervised SPL-MAD solution outperforms the overall performance of a wide range of supervised MAD solutions and provides higher generalizability on unknown attacks. Training codes and pre-trained models are publicly released11https://github.com/meilfang/SPL-MAD.

This paper presents a summary of the Competition on Face Morphing Attack Detection Based on Privacy-aware Synthetic Training Data (SYN-MAD) held at the 2022 In-ternational Joint Conference on Biometrics (IJCB 2022). The competition attracted a total of 12 participating teams, both from academia and industry and present in 11 differ-ent countries. In the end, seven valid submissions were submitted by the participating teams and evaluated by the organizers. The competition was held to present and at-tract solutions that deal with detecting face morphing at-tacks while protecting people's privacy for ethical and le-gal reasons. To ensure this, the training data was limited to synthetic data provided by the organizers. The submitted solutions presented innovations that led to out-performing the considered baseline in many experimental settings. The evaluation benchmark is now available at: https://github.com/marcohuber/SYN-MAD-2022.

Deep learning-based systems for periocular recognition make use of the high recognition performance of neural networks, which, however, is accompanied by high computational costs and memory footprints. This can lead to deployability problems, especially in mobile devices and embedded systems. Few previous works strived towards building lighter models, however, while still depending on floating-point numbers associated with higher computational cost and memory footprint. In this paper, we propose to adapt model quantization for periocular recognition. This, within the proposed scheme, leads to reducing the memory footprint of periocular recognition network by up to five folds while maintaining high recognition performance. We present a comprehensive analysis over three backbones and diverse experimental protocols to stress the consistency of our conclusions, along with a comparison with a wide set of baselines that prove the optimal trade-off between performance and model size achieved by our proposed solution. The code and pre-trained models have been made available at https://github.com/jankolf/ijcb-periocular-quantization.

This work summarizes the IJCB Occluded Face Recognition Competition 2022 (IJCB-OCFR-2022) embraced by the 2022 International Joint Conference on Biometrics (IJCB 2022). OCFR-2022 attracted a total of 3 participating teams, from academia. Eventually, six valid submissions were submitted and then evaluated by the organizers. The competition was held to address the challenge of face recognition in the presence of severe face occlusions. The participants were free to use any training data and the testing data was built by the organisers by synthetically occluding parts of the face images using a well-known dataset. The submitted solutions presented innovations and performed very competitively with the considered baseline. A major output of this competition is a challenging, realistic, and diverse, and publicly available occluded face recognition benchmark with well defined evaluation protocols.

Deep learning-based face recognition models follow the common trend in deep neural networks by utilizing full-precision floating-point networks with high computational costs. Deploying such networks in use-cases constrained by computational requirements is often infeasible due to the large memory required by the full-precision model. Previous compact face recognition approaches proposed to design special compact architectures and train them from scratch using real training data, which may not be available in a real-world scenario due to privacy concerns. We present in this work the QuantFace solution based on low-bit precision format model quantization. QuantFace reduces the required computational cost of the existing face recognition models without the need for designing a particular architecture or accessing real training data. QuantFace introduces privacy-friendly synthetic face data to the quantization process to mitigate potential privacy concerns and issues related to the accessibility to real training data. Through extensive evaluation experiments on seven benchmarks and four network architectures, we demonstrate that QuantFace can successfully reduce the model size up to 5x while maintaining, to a large degree, the verification performance of the full-precision model without accessing real training datasets. All training codes are publicly available1.

Abstract Iris Presentation Attack Detection (PAD) algorithms address the vulnerability of iris recognition systems to presentation attacks. With the great success of deep learning methods in various computer vision fields, neural network-based iris PAD algorithms emerged. However, most PAD networks suffer from overfitting due to insufficient iris data variability. Therefore, we explore the impact of various data augmentation techniques on performance and the generalizability of iris PAD. We apply several data augmentation methods to generate variability, such as shift, rotation, and brightness. We provide in-depth analyses of the overlapping effect of these methods on performance. In addition to these widely used augmentation techniques, we also propose an augmentation selection protocol based on the assumption that various augmentation techniques contribute differently to the PAD performance. Moreover, two fusion methods are performed for more comparisons: the strategy-level and the score-level combination. We demonstrate experiments on two fine-tuned models and one trained from the scratch network and perform on the datasets in the Iris-LivDet-2017 competition designed for generalizability evaluation. Our experimental results show that augmentation methods improve iris PAD performance in many cases. Our least overlap-based augmentation selection protocol achieves the lower error rates for two networks. Besides, the shift augmentation strategy also exceeds state-of-the-art (SoTA) algorithms on the Clarkson and IIITD-WVU datasets.

Wearing a mask has proven to be one of the most effective ways to prevent the transmission of SARS-Co V-2 coronavirus. However, wearing a mask poses challenges for different face recognition tasks and raises concerns about the performance of masked face presentation detection (PAD). The main issues facing the mask face PAD are the wrongly classified bona fide masked faces and the wrongly classified partial attacks (covered by real masks). This work addresses these issues by proposing a method that considers partial attack labels to supervise the PAD model training, as well as regional weighted inference to further improve the PAD performance by varying the focus on different facial areas. Our proposed method is not directly linked to specific network architecture and thus can be directly incorporated into any common or custom-designed network. In our work, two neural networks (DeepPixBis [21] and MixFaceNet [4]) are selected as backbones. The experiments are demonstrated on the collaborative real mask attack (CRMA) database [17]. Our proposed method outperforms established PAD methods in the CRMA database by reducing the mentioned shortcomings when facing masked faces. Moreover, we present a detailed step-wise ablation study pointing out the individual and joint benefits of the proposed concepts on the overall PAD performance.

The emergence of the global COVID-19 pandemic poses new challenges for biometrics. Not only are contactless biometric identification options becoming more important, but face recognition has also recently been confronted with the frequent wearing of masks. These masks affect the performance of previous face recognition systems, as they hide important identity information. In this paper, we propose a mask-invariant face recognition solution (MaskInv) that utilizes template-level knowledge distillation within a training paradigm that aims at producing embeddings of masked faces that are similar to those of non-masked faces of the same identities. In addition to the distilled knowledge, the student network benefits from additional guidance by margin-based identity classification loss, ElasticFace, using masked and non-masked faces. In a step-wise ablation study on two real masked face databases and five mainstream databases with synthetic masks, we prove the rationalization of our MaskInv approach. Our proposed solution outperforms previous state-of-the-art (SOTA) academic solutions in the recent MFRC-21 challenge in both scenarios, masked vs masked and masked vs non-masked, and also outperforms the previous solution on the MFR2 dataset. Furthermore, we demonstrate that the proposed model can still perform well on unmasked faces with only a minor loss in verification performance. The code, the trained models, as well as the evaluation protocol on the synthetically masked data are publicly available: https://github.com/fdbtrs/Masked-Face-Recognition-KD.

SARS-CoV-2 has presented direct and indirect challenges to the scientific community. One of the most prominent indirect challenges advents from the mandatory use of face masks in a large number of countries. Face recognition methods struggle to perform identity verification with similar accuracy on masked and unmasked individuals. It has been shown that the performance of these methods drops considerably in the presence of face masks, especially if the reference image is unmasked. We propose FocusFace, a multi-task architecture that uses contrastive learning to be able to accurately perform masked face recognition. The proposed architecture is designed to be trained from scratch or to work on top of state-of-the-art face recognition methods without sacrificing the capabilities of a existing models in conventional face recognition tasks. We also explore different approaches to design the contrastive learning module. Results are presented in terms of masked-masked (M-M) and unmasked-masked (U-M) face verification performance. For both settings, the results are on par with published methods, but for M-M specifically, the proposed method was able to outperform all the solutions that it was compared to. We further show that when using our method on top of already existing methods the training computational costs decrease significantly while retaining similar performances. The implementation and the trained models are available at GitHub.

In this paper, we present a set of extremely efficient and high throughput models for accurate face verification, Mix-FaceNets which are inspired by Mixed Depthwise Convolutional Kernels. Extensive experiment evaluations on Label Face in the Wild (LFW), Age-DB, MegaFace, and IARPA Janus Benchmarks IJB-B and IJB-C datasets have shown the effectiveness of our MixFaceNets for applications requiring extremely low computational complexity. Under the same level of computation complexity (≤ 500M FLOPs), our MixFaceNets outperform MobileFaceNets on all the evaluated datasets, achieving 99.60% accuracy on LFW, 97.05% accuracy on AgeDB-30, 93.60 TAR (at FAR1e-6) on MegaFace, 90.94 TAR (at FAR1e-4) on IJB-B and 93.08 TAR (at FAR1e-4) on IJB-C. With computational complexity between 500M and 1G FLOPs, our MixFaceNets achieved results comparable to the top-ranked models, while using significantly fewer FLOPs and less computation over-head, which proves the practical value of our proposed Mix-FaceNets. All training codes, pre-trained models, and training logs have been made available https://github.com/fdbtrs/mixfacenets.

This paper presents a summary of the Masked Face Recognition Competitions (MFR) held within the 2021 International Joint Conference on Biometrics (IJCB 2021). The competition attracted a total of 10 participating teams with valid submissions. The affiliations of these teams are diverse and associated with academia and industry in nine different countries. These teams successfully submitted 18 valid solutions. The competition is designed to motivate solutions aiming at enhancing the face recognition accuracy of masked faces. Moreover, the competition considered the deployability of the proposed solutions by taking the compactness of the face recognition models into account. A private dataset representing a collaborative, multisession, real masked, capture scenario is used to evaluate the submitted solutions. In comparison to one of the topperforming academic face recognition solutions, 10 out of the 18 submitted solutions did score higher masked face verification accuracy.

Iris presentation attack detection (PAD) plays a vital role in iris recognition systems. Most existing CNN-based iris PAD solutions 1) perform only binary label supervision during the training of CNNs, serving global information learning but weakening the capture of local discriminative features, 2) prefer the stacked deeper convolutions or expert-designed networks, raising the risk of overfitting, 3) fuse multiple PAD systems or various types of features, increasing difficulty for deployment on mobile devices. Hence, we propose a novel attention-based deep pixel-wise bi-nary supervision (A-PBS) method. Pixel-wise supervision is first able to capture the fine-grained pixel/patch-level cues. Then, the attention mechanism guides the network to automatically find regions that most contribute to an accurate PAD decision. Extensive experiments are performed on LivDet-Iris 2017 and three other publicly available databases to show the effectiveness and robustness of proposed A-PBS methods. For instance, the A-PBS model achieves an HTER of 6.50% on the IIITD-WVU database outperforming state-of-the-art methods.

Liveness Detection (LivDet)-Face is an international competition series open to academia and industry. The competition’s objective is to assess and report state-of-the-art in liveness / Presentation Attack Detection (PAD) for face recognition. Impersonation and presentation of false samples to the sensors can be classified as presentation attacks and the ability for the sensors to detect such attempts is known as PAD. LivDet-Face 2021 * will be the first edition of the face liveness competition. This competition serves as an important benchmark in face presentation attack detection, offering (a) an independent assessment of the current state of the art in face PAD, and (b) a common evaluation protocol, availability of Presentation Attack Instruments (PAI) and live face image dataset through the Biometric Evaluation and Testing (BEAT) platform. The competition can be easily followed by researchers after it is closed, in a platform in which participants can compare their solutions against the LivDet-Face winners.

For iris recognition in non-cooperative environments, iris segmentation has been regarded as the first most important challenge still open to the biometric community, affecting all downstream tasks from normalization to recognition. In recent years, deep learning technologies have gained significant popularity among various computer vision tasks and also been introduced in iris biometrics, especially iris segmentation. To investigate recent developments and attract more interest of researchers in the iris segmentation method, we organized the 2021 NIR Iris Challenge Evaluation in Non-cooperative Environments: Segmentation and Localization (NIR-ISL 2021) at the 2021 International Joint Conference on Biometrics (IJCB 2021). The challenge was used as a public platform to assess the performance of iris segmentation and localization methods on Asian and African NIR iris images captured in non-cooperative environments. The three best-performing entries achieved solid and satisfactory iris segmentation and localization results in most cases, and their code and models have been made publicly available for reproducibility research.

The growing popularity of Virtual Reality and Augmented Reality (VR/AR) devices in many applications also demands authentication of users. As the devices inherently capture the eye image while capturing the user interaction, the authentication can be devised using the iris and periocular recognition. While both iris and periocular data being non-ideal unlike the data captured from standard biometric sensors, the authentication performance is expected to be lower. In this work, we present and evaluate a fusion framework for improving the biometric authentication performance. Specifically, we employ score-level fusion for two independent biometric systems of iris and periocular region to avoid expensive feature-level fusion. With a detailed evaluation of three different score-level fusion after the score normalization on a dataset of 12579 images, we report the performance gain in authentication using score-level fusion for iris and periocular recognition.

Iris presentation attack detection (PAD) algorithms are developed to address the vulnerability of iris recognition systems to presentation attacks. Taking into account that the deep features successfully improved computer vision performance in various fields including iris recognition, it is natural to use features extracted from deep neural networks for iris PAD. Each layer in a deep learning network carries features of different level of abstraction. The features extracted from the first layer to the higher layers become more complex and more abstract. This might point our complementary information in these features that can collaborate towards an accurate PAD decision. Therefore, we propose an iris PAD solution based on multi-layer fusion. The information extracted from the last several convolutional layers are fused on two levels, feature-level and score-level. We demonstrated experiments on both, off-the-shelf pre-trained network and network trained from scratch. An extensive experiment also explores the complementary between different layer combinations of deep features. Our experimental results show that feature-level based multi-layer fusion method performs better than the best single layer feature extractor in most cases. In addition, our fusion results achieve similar or better results than the state-of-the-art algorithms on the Notre Dame and IIITD-WVU databases of the Iris Liveness Detection Competition 2017 (LivDet-Iris 2017).

Augmented and virtual reality is being deployed in different fields of applications. Such applications might involve accessing or processing critical and sensitive information, which requires strict and continuous access control. Given that Head-Mounted Displays (HMD) developed for such applications commonly contains internal cameras for gaze tracking purposes, we evaluate the suitability of such setup for verifying the users through iris recognition. In this work, we first evaluate a set of iris recognition algorithms suitable for HMD devices by investigating three well-established handcrafted feature extraction approaches, and to complement it, we also present the analysis using four deep learning models. While taking into consideration the minimalistic hardware requirements of stand-alone HMD, we employ and adapt a recently developed miniature segmentation model (EyeMMS) for segmenting the iris. Further, to account for non-ideal and non-collaborative capture of iris, we define a new iris quality metric that we termed as Iris Mask Ratio (IMR) to quantify the iris recognition performance. Motivated by the performance of iris recognition, we also propose the continuous authentication of users in a non-collaborative capture setting in HMD. Through the experiments on a publicly available OpenEDS dataset, we show that performance with EER = 5% can be achieved using deep learning methods in a general setting, along with high accuracy for continuous user authentication.

Launched in 2013, LivDet-Iris is an international competition series open to academia and industry with the aim to assess and report advances in iris Presentation Attack Detection (PAD). This paper presents results from the fourth competition of the series: LivDet-Iris 2020. This year's competition introduced several novel elements: (a) incorporated new types of attacks (samples displayed on a screen, cadaver eyes and prosthetic eyes), (b) initiated LivDet-Iris as an on-going effort, with a testing protocol available now to everyone via the Biometrics Evaluation and Testing (BEAT)* open-source platform to facilitate reproducibility and benchmarking of new algorithms continuously, and (c) performance comparison of the submitted entries with three baseline methods (offered by the University of Notre Dame and Michigan State University), and three open-source iris PAD methods available in the public domain. The best performing entry to the competition reported a weighted average APCER of 59.10% and a BPCER of 0.46% over all five attack types. This paper serves as the latest evaluation of iris PAD on a large spectrum of presentation attack instruments.

The paper presents a summary of the 2020 Sclera Segmentation Benchmarking Competition (SSBC), the 7th in the series of group benchmarking efforts centred around the problem of sclera segmentation. Different from previous editions, the goal of SSBC 2020 was to evaluate the performance of sclera-segmentation models on images captured with mobile devices. The competition was used as a platform to assess the sensitivity of existing models to i) differences in mobile devices used for image capture and ii) changes in the ambient acquisition conditions. 26 research groups registered for SSBC 2020, out of which 13 took part in the final round and submitted a total of 16 segmentation models for scoring. These included a wide variety of deep-learning solutions as well as one approach based on standard image processing techniques. Experiments were conducted with three recent datasets. Most of the segmentation models achieved relatively consistent performance across images captured with different mobile devices (with slight differences across devices), but struggled most with low-quality images captured in challenging ambient conditions, i.e., in an indoor environment and with poor lighting.