Lightweight Deep Learning Models for Face Mask Detection in Real-Time Edge Environments: A Review and Future Research Directions

1. Introduction

The widespread proliferation of airborne infectious diseases such as COVID-19 highlighted the importance of face coverings as an effective measure for reducing person-to-person transmission in both clinical and community environments [1]. However, the relevance of automated mask detection extends well beyond pandemic response. In many occupational settings, including healthcare, pharmaceutical laboratories, hospitality services, manufacturing plants, and food-handling industries, mask-wearing remains a mandatory hygiene and safety requirement [2]. In such environments, inconsistent or improper mask usage can compromise safety, reduce operational compliance, and increase the risk of contamination. Automated, computer-vision-based monitoring systems therefore provide meaningful value by detecting mask-wearing violations and triggering real-time alerts to reduce reliance on manual supervision [2,3].

Deep learning-based face mask detection has consequently emerged as a scalable, non-intrusive solution for compliance monitoring across public and professional domains [2,3]. As artificial intelligence (AI) increasingly shifts toward on-device and edge-computing environments, research emphasis has expanded from merely achieving high accuracy to achieving high accuracy with real-time efficiency under hardware limitations [4,5]. This shift reframes the central research question from “How do we detect masks accurately?” to “How do we detect masks accurately and efficiently enough for deployment on constrained edge platforms?”

Traditional CNN architectures such as VGGNet and ResNet deliver strong feature-learning capabilities but are computationally intensive and often unsuitable for deployment on low-power devices. Lightweight architectures, including the MobileNet family and other efficient convolutional models, address these challenges by reducing parameter count, lowering latency, and enabling deployment on embedded systems [3]. More recently, hybrid architectures that integrate lightweight feature extractors with optimized one-stage detectors (e.g., SSD-, YOLO-, or attention-enhanced variants) have demonstrated promising improvements in balancing accuracy, speed, and resource usage [5,6]. These advancements suggest that robust and efficient mask detection is feasible even within strict real-time constraints.

Problem Statement and Rationale: Despite considerable progress, practical deployment of mask detection systems continues to face two major challenges:

(1) maintaining high recognition reliability under real-world variability such as occlusion, lighting changes, and diverse mask types, and

(2) operating efficiently on hardware-limited platforms where computational resources, memory, and energy budgets are restricted [4,5].

Lightweight and hybrid architectures offer promising solutions, yet the rapid diversification of efficient model designs has created uncertainty regarding which architectural choices best balance accuracy, inference speed, and deployment feasibility. To address this gap, this review synthesizes contemporary architectural approaches for mask detection, compares their performance characteristics, and identifies research opportunities to guide future development [6,7,8,9].

Existing surveys and review efforts remain limited in several respects. Many emphasize pandemic-driven solutions without considering broader compliance applications beyond COVID-19 [7,8,9]. Others focus heavily on accuracy while giving insufficient attention to practical factors such as latency, model size, generalization capacity, or the constraints of embedded deployment [4,5,8]. Issues related to improper mask detection, domain shift, and long-term sustainability remain largely unresolved, and emerging methods for compression, distillation, and hardware-aware optimization are underexplored in current literature. This review aims to address these gaps by providing an architecture-centric analysis followed by a performance comparison and a forward-looking discussion of open challenges.

This review aims to:

a) Examine conventional, lightweight, and hybrid deep learning architectures used for face mask detection;

b) Compare their performance with respect to accuracy, inference efficiency, and deployment suitability;

c) Analyze the core challenges affecting real-world deployment, including improper mask detection, domain shift, and computational constraints;

d) Outline future research directions focused on model compression, knowledge distillation, domain adaptation, and expanding applications beyond mask detection.

The remainder of this paper is organized as follows. Section 2 presents the methodology adopted for conducting the review. Section 3 analyzes the major architectural families employed in face mask detection systems. Section 4 provides a comparative evaluation of model performance, focusing on accuracy and efficiency considerations. Section 5 discusses the open challenges and outlines potential directions for future research. Finally, Section 6 concludes the paper by summarizing the key findings and contributions.

2. Methodology

The methodology of this review was designed to rigorously analyze lightweight and hybrid deep learning approaches for face mask detection, with particular emphasis on their feasibility in real-time and resource-constrained deployment environments. The assessment focused not only on reported performance metrics such as accuracy and inference speed, but also on architectural efficiency, evaluation settings, hardware compatibility, and scalability. Rather than following a rigid systematic protocol, the selection and evaluation process were guided by conceptual relevance and alignment with the research objectives defined earlier. Each included study was examined through a multi-dimensional analytical framework considering the problem addressed, the proposed architecture, optimization strategies, dataset characteristics, experimental setup, reported strengths and limitations, and potential research gaps.

2.1. Literature Search Strategy

The literature search was conducted using peer-reviewed digital libraries including ScienceDirect, IEEE Xplore, SpringerLink, ACM Digital Library, and MDPI. Publications from 2020 to 2025 were considered to capture developments aligned with post-pandemic transitions and emerging edge-based AI deployment trends. Search terms included combinations of “face mask detection,” “lightweight deep learning,” “real-time inference,” “edge computing,” “MobileNet,” “YOLO,” “improper mask detection,” “embedded deployment,” and “deep learning optimization.” Only journal articles and reputable conference proceedings indexed in SCI, SCIE, or Scopus were shortlisted. Reference tracing was additionally employed to capture studies with architectural innovation or deployment-oriented experimentation.

2.2. Inclusion and Exclusion Criteria

Studies were included if they presented a deep learning–based method for face mask detection (binary or multi-class) and provided quantitative evaluation using publicly available or well-described datasets. Priority was given to research addressing real-time inference, constrained-resource deployment, architectural efficiency, or model-level optimization. Studies without empirical experimentation, those focused exclusively on cloud-based inference without discussion of deployment feasibility, and works based solely on traditional image processing were excluded. Non-peer-reviewed preprints and duplicates were also omitted to ensure reliability.

2.3. Screening and Selection Approach

Screening was conducted in multiple stages. Initial filtering based on titles and abstracts ensured alignment with lightweight architectures and deployment-aware mask detection. Full-text screening was then used to assess architectural contributions, optimization strategies, dataset suitability for the stated task, and real-time implementation potential. Articles that described mask detection without addressing computational or deployment constraints were retained only as secondary background sources rather than core analytical studies.

2.4. Data Extraction and Categorization

For each selected study, key information was extracted to support thematic and architectural analysis. This included the research problem, architectural design, enhancement strategies, dataset and evaluation conditions, training setup, performance metrics, model complexity (e.g., parameter count), inference performance (e.g., FPS), and any deployment-specific details such as hardware specifications or optimization pipelines. Studies were categorized into conventional CNNs, lightweight convolutional models, and hybrid architectures. This structured categorization enabled comparative analysis of architectural trends, efficiency trade-offs, and practical deployment implications discussed in later sections. To ensure transparency and consistency in the comparative review, the essential characteristics of each included study were extracted and organized into a structured summary. This overview captures the goals, methodological designs, experimental settings, and principal findings of the selected works. Table 1 presents these extracted details and provides a foundational context for the architectural and performance analyses discussed in subsequent sections.

The studies summarized in Table 1 reveal significant variation in architectural choices, dataset configurations, evaluation procedures, and deployment assumptions. This diversity underscores the need for an architecture-centric viewpoint when interpreting performance results, as models differ not only in structural complexity but also in experimental conditions and optimization strategies. The extracted patterns also highlight gaps, such as inconsistent reporting of inference metrics, limited multi-class improper mask annotations, and scarce evaluation under real-world deployment constraints that motivate the architectural comparison and performance analysis presented in upcoming sections. These observations further justify the focus of this review on understanding how architectural families influence efficiency, robustness, and readiness for edge deployment.

3. Architectural Landscape of Face Mask Detection Models

Deep learning–based face mask detection systems rely fundamentally on the architectural design of their underlying neural networks. Architectural choices determine not only a model’s feature-extraction capacity and recognition accuracy but also its computational footprint, memory usage, inference speed, and overall suitability for deployment on resource-constrained edge environments. As mask detection has transitioned from high-performance computing systems toward embedded and real-time monitoring platforms, the efficiency and scalability of these architectures have become just as important as their classification accuracy.

This section provides an architecture-oriented review of face mask detection models, examining how different architectural families address the competing demands of accuracy and efficiency. The discussion progresses from conventional convolutional neural networks, which offer strong representational power but high computational cost, to lightweight convolutional models specifically optimized for mobile and embedded inference. The section then evaluates hybrid architectures that integrate lightweight backbones with streamlined detection heads or specialized optimization strategies to achieve real-time performance without substantial loss of accuracy. Together, these architectural categories reflect the evolution of face mask detection from computationally expensive methods toward edge-ready solutions suitable for widespread deployment.

3.1. Conventional CNN-Based Approaches

Early work on automated face mask detection relied heavily on conventional convolutional neural network (CNN) architectures, which formed the foundation of modern deep learning in computer vision. These models, originally developed for large-scale image classification and object detection tasks, were adopted due to their strong representational capacity and readily available pre-trained weights. Architectures such as AlexNet, VGGNet, Inception, ResNet, DenseNet, Xception, and the R-CNN family provided the initial backbone for mask detection pipelines during the early phase of the COVID-19 pandemic.

The evolution of conventional CNNs reflects a gradual improvement in depth, efficiency, and stability. LeNet-5 introduced the first successful convolutional network structure for digit recognition, demonstrating the feasibility of learned hierarchical features [16]. AlexNet catalyzed the deep learning revolution by winning the ImageNet 2012 challenge, leveraging ReLU activations and GPU training to drastically outperform traditional hand-crafted descriptors [17]. VGG16/VGG19 refined network depth with small, uniform 3×3 convolutions, producing strong yet computationally heavy models [18]. The Inception family introduced multi-branch processing to improve efficiency while preserving expressive power [19], while Inception-ResNet blended residual learning with inception modules to enable deeper and more stable optimization [20].

Several major CNN families subsequently introduced important structural innovations. ResNet, with its identity skip connections, addressed the vanishing-gradient problem and enabled successful training of networks exceeding 100 layers [21]. DenseNet extended this concept by connecting each layer to all subsequent layers, improving feature reuse and reducing parameter growth [22]. Xception reorganized convolutions into fully depthwise separable form, a precursor to later lightweight models such as MobileNet [23]. More recent backbone families such as RegNet attempted to design regular, scalable architectures via design-space exploration, though these were not widely adopted for mask detection because of computational demands [24].

Beyond classification backbones, conventional object detection frameworks also played a significant role. The R-CNN family including R-CNN, Fast R-CNN, and Faster R-CNN, extended CNNs to region-based object detection using two-stage processing [25,26,27]. Mask R-CNN further introduced an instance segmentation branch, enabling pixel-level analysis of facial regions or mask boundaries [28]. These models were frequently adopted in early COVID-19 surveillance systems, especially in industrial or controlled environments requiring both face detection and classification.

In the context of face-mask detection, conventional CNN backbones were widely adopted in early studies implementing either face-classification pipelines or end-to-end object-detection frameworks. A typical classification workflow included: (i) frame acquisition from CCTV or mobile cameras, (ii) pre-processing such as resizing and normalization, (iii) face detection using Haar cascades, HOG-based detectors, or CNN-based region detectors, and (iv) classification of each cropped face using a CNN such as VGG16, ResNet50, DenseNet121 or InceptionV3. Also, conventional CNNs and detection frameworks were employed early on to evaluate the feasibility of automated mask compliance. For instance, MaskedFace-Net, a large publicly available dataset containing over 137,000 images with correct and incorrect mask wearing annotations, was used to train and test classifiers based on VGG, ResNet, or DenseNet backbones; results demonstrated that these models could reliably distinguish between masked and unmasked faces under controlled image conditions [29].

Conventional CNN backbones have been employed in both face-classification and object-detection pipelines for mask detection. For instance, the MaskedFace-Net dataset, containing over 137,000 images with masks correctly or incorrectly worn, has served to train classification models using architectures such as VGG, ResNet or DenseNet, showing that conventional CNNs can reliably distinguish between masked and unmasked faces under controlled conditions [29]. On the detection side, Jiang et al. proposed SE-YOLOv3, a YOLOv3-based detector enhanced with a Squeeze-and-Excitation module, trained on the PWMFD dataset of 9,205 images. Their results indicated high detection accuracy along with real-time inference capability, demonstrating that conventional CNN backbones remain effective when combined with modern detection heads [30].

However, performance of conventional CNN-based detectors degrades under real-world variability. A recent survey on masked-face recognition and detection notes that occlusion (e.g., by hands or non-mask objects), non-frontal poses, diverse mask designs, and inconsistent lighting conditions significantly reduce the reliability of classification and detection models based on conventional CNNs [31]. Even large annotated datasets like MaskedFace-Net, though extensive but often contain mostly frontal or simply synthesized mask images, limiting their representativeness for unconstrained environments [29].

Overall, while conventional CNN architectures continue to provide a valuable baseline for face-mask detection tasks, their robustness and generalization remain limited in challenging, real-world scenarios. The observed deficiencies under occlusion, pose variation, and diverse mask usage conditions motivate the shift toward lightweight CNNs, optimized detection heads, and hybrid architectures, which strive to balance accuracy with computational efficiency and deployment feasibility in edge or surveillance environments.

A comparative overview of major conventional CNN architectures, highlighting their innovations, parameter scales and deployment implications, is presented in Table 2.

Despite their strengths, these networks exhibit important limitations for real-time deployment and motivated the rise of lightweight CNNs (e.g., MobileNet, ShuffleNet) and hybrid architectures designed for edge-optimized deployment while retaining adequate accuracy.

3.2. Lightweight Convolutional Models

Lightweight convolutional neural networks are designed to provide competitive recognition accuracy while significantly reducing memory footprint, parameter count and computational cost. Instead of relying on wide and deep stacks of standard convolutions, these models adopt mechanisms such as depth-wise separable convolutions, inverted residual bottlenecks, grouped pointwise convolutions with channel shuffling, or squeeze–expand fire modules to minimize the number of multiplications and parameters. As a result, they are well suited for deployment on mobile and embedded platforms, including edge devices used in camera-based mask monitoring systems.

Among these architectures, the MobileNet family is arguably the most influential. The original MobileNet formulation introduced a streamlined architecture based on depth-wise separable convolutions and global width and resolution multipliers, enabling a tunable trade-off between latency and accuracy for mobile vision tasks [32]. MobileNetV2 extended this idea by incorporating inverted residual blocks with linear bottlenecks, improving accuracy across multiple benchmarks while preserving efficiency [33]. Later variants such as MobileNetV3 [34] further refined block design and activation functions, but MobileNetV2 remains the most widely adopted backbone in the face mask detection literature due to its balance between representational power and computational cost.

Other lightweight architectures follow different optimization strategies. EfficientNet introduced compound scaling of depth, width and input resolution, combined with an architecture discovered via neural architecture search, producing a family of models that deliver state-of-the-art accuracy with substantially fewer parameters than conventional CNNs [11]. ShuffleNet instead reduces computational cost by using pointwise group convolutions and channel shuffle, allowing efficient information mixing while maintaining high throughput on ARM-based mobile devices [35]. SqueezeNet targets extreme parameter reduction by replacing many standard convolution layers with fire modules (1×1 squeeze layers followed by 1×1 and 3×3 expand layers), achieving AlexNet-level ImageNet accuracy with roughly 50× fewer parameters [36]. These architectures form the algorithmic foundation upon which many recent lightweight mask detection systems are built.

In the specific context of face mask detection, MobileNetV2 is the most frequently used lightweight backbone. Several studies employ MobileNetV2, either as a feature extractor within an object-detection pipeline or as the main classifier in a transfer-learning setting. For example, one real-time system (SSDMNV2) integrates MobileNetV2 into a single-shot detection framework and reports effective two-class (mask/no-mask) detection using OpenCV, TensorFlow and Keras on live video streams [37]. Other works fine-tune MobileNetV2 on custom mask datasets and then attach classical machine-learning classifiers (such as SVM or decision trees) on top of deep features, demonstrating that MobileNetV2-based embeddings can be reused across a variety of deployment strategies [38]. Additional implementations exploit MobileNetV2 in embedded or IoT-oriented systems to enforce mask-wearing compliance, emphasizing its ability to achieve real-time inference on constrained hardware while maintaining high recognition accuracy [39].

EfficientNet-based models have also been explored for mask detection, typically as stronger yet still relatively compact alternatives to MobileNet. One study employing EfficientNet-B0 for binary mask classification reports an accuracy of around 99–99.7% on a two-class problem, outperforming several heavier backbones while remaining deployable in practical systems [40]. In [41] the authors employ EfficientNet-B0 as the feature extractor backbone and a large-margin piecewise-linear (LMPL) classifier on top of deep features. The method achieved 99.53% and 99.64% accuracies for the two tasks respectively, outperforming conventional end-to-end CNN models and classical image classification approaches. The authors argue that their solution achieves both high accuracy and efficiency, making it suitable for real-world deployment in scenarios like biometric authentication or public-health compliance checks. Other works using EfficientNet variants for mask identification similarly highlight their favorable accuracy–efficiency trade-off compared with legacy CNNs, though their computational requirements are generally higher than those of MobileNetV2 on extremely low-power devices [42].

Architectures derived from SqueezeNet provide another lightweight option. SqueezeNet itself was designed as an extremely compact classification network, but face mask detection has inspired specialized derivatives such as SqueezeMaskNet, which integrates fire modules with channel-attention mechanisms to support four-way classification (correctly worn mask, mask covering only the mouth, mask not covering, and no mask) while running in real time on edge GPUs [43]. Comparative evaluations indicate that SqueezeMaskNet achieves accuracy in the mid- to high-90% range while sustaining high frame rates on Jetson-class hardware, making it particularly suitable for embedded applications where classical CNNs would be too heavy.

Finally, several hybrid lightweight architectures combine these backbones with efficient detection heads. For instance, MobileNetV2 features can be paired with SSD-style detection layers or YOLO derivatives, and EfficientNet or SqueezeNet variants can be integrated into multi-stage pipelines that first detect faces and then classify mask usage [37]. Survey studies on mask detection also confirm that MobileNetV2-based solutions dominate the lightweight category, with EfficientNet and SqueezeNet-derived models appearing as strong but less commonly used alternatives [8]. Overall, the literature suggests that MobileNetV2 is currently the most popular lightweight model for face mask detection, while EfficientNet and SqueezeMaskNet offer accuracy gains in applications where slightly higher computational budgets are acceptable. A structured comparison of these lightweight architectures is presented in Table 3 to highlight their respective design strategies, computational efficiency and suitability for real-time face-mask detection

Lightweight architectures have significantly improved the feasibility of real-time face-mask detection on embedded platforms by minimizing computational overhead without substantially compromising recognition accuracy. Among these models, MobileNetV2 remains the most commonly deployed due to its favorable balance between architectural simplicity and inference efficiency, particularly when used in transfer-learning pipelines. EfficientNet-based implementations, such as EfficientMask-Net, demonstrate superior accuracy, especially for improper mask detection, though their computational demands are marginally higher. SqueezeMaskNet, derived from SqueezeNet with channel-attention enhancements, offers exceptional frame rates on edge GPUs and supports four-class decision schemes, making it suitable for rapid compliance verification. ShuffleNet is rarely preferred due to inconsistent performance, whereas hybrid configurations integrating MobileNet or EfficientNet with YOLO-like detection heads provide strong localization capabilities for surveillance applications. Overall, MobileNetV2 is most widely adopted, while EfficientNet derivatives lead in accuracy-sensitive tasks, and SqueezeMaskNet offers the best real-time responsiveness in constrained environments.

3.3. Hybrid Architectures

Hybrid architecture represents an advanced design philosophy that extends beyond standalone conventional CNNs or lightweight backbones by combining multiple complementary modules, typically a feature extraction network and a detection or classification head that are used to handle both spatial localization and mask classification effectively. Unlike the approaches in Section 3.1 and Section 3.2, hybrid architectures integrate networks such as MobileNetV2, EfficientNet, or SqueezeNet with detection frameworks like YOLO, SSD, or Faster R-CNN, or with classical ML classifiers. These combinations are not random; they arise from deliberate architectural reasoning to meet real-time deployment demands, especially in public health monitoring during COVID-19. Researchers have demonstrated that hybridization improves robustness, speed, and accuracy under real-world constraints, such as occlusion, inconsistent lighting and crowd density (e.g., YOLOv2+ResNet50 [44], MobileNetV2+SSD [45], YOLOv5 with attention mechanisms [46], or CNN+SVM hybrids [47]).

Lightweight hybrid architectures are crucial for face mask detection, especially in the context of the COVID-19 pandemic, where real-time monitoring in public spaces is essential for public health compliance. These architectures enable deployment on edge devices, which are often resource-constrained, by reducing computational costs and memory requirements while maintaining high detection accuracy. The integration of lightweight models such as MobileNetV2, and the use of heavier feature extractors like VGG19 within hybrid pipelines, allows for efficient feature extraction and classification, making them suitable for real-time applications in environments with limited hardware capabilities [48,49,50,51,52].

This motivation aligns with findings from multiple hybrid studies, where integrating lightweight backbones with advanced detection heads significantly improved end-to-end performance for mask localization and classification in constrained settings (e.g., YOLOv3-based hybrids [53], ensemble hybrid detectors [2], and smart-city hybrid systems[54]).

The design of lightweight hybrid architectures typically involves combining the strengths of different model types, such as convolutional neural networks (CNNs) and transformer models, to optimize both efficiency and accuracy. Techniques like depth-wise separable convolutions, inverted residual structures, and parallel hybrid architectures are commonly employed to minimize the number of parameters and computational operations. These principles ensure that models can operate effectively on mobile and edge devices without sacrificing performance [48,49,51,55].

Such design principles are evident in real hybrid implementations. For example, YOLOv5 combined with coordinate attention blocks improves detection precision while maintaining fast inference[46], and CNN+ML hybrids (deep features + SVM) exploit efficiency while preserving discriminative power [47]. These works demonstrate that hybridization follows structured engineering choices rather than arbitrary pairing of networks.

Developers of lightweight hybrid architecture must balance the trade-off between computational efficiency and detection accuracy. While lightweight models are optimized for speed and resource usage, they may experience accuracy degradation under challenging conditions such as illumination changes or occlusions. Hybrid models attempt to mitigate these issues by leveraging complementary strengths of different architectures, but some loss in robustness may still occur in extreme scenarios [56,57,58].

This trade-off is well-documented in comparison studies, where hybrid detectors show better occlusion-robustness than pure classifiers. For instance, YOLOv2-ResNet50 hybrids handle masked and partially occluded faces more effectively than standalone CNNs [44], and multi-detector ensembles enhance detection under crowd conditions [2]. Nevertheless, no hybrid fully eliminates robustness issues, especially under severe occlusions or rapid motion.

To further enhance the performance of lightweight hybrid models, various optimization techniques are applied. These include model compression, quantization, pruning, and knowledge distillation, all aimed at reducing model size and computational demands. Additionally, architectural innovations such as ghost modules and Bi-FPN (Bidirectional Feature Pyramid Networks) are explored to improve feature extraction and detection capabilities without increasing resource consumption [51,57].

These optimization strategies also appear in hybrid architectures that rely on TensorRT acceleration, auto-labelling modules, and attention-based feature refinement, as seen in improved YOLOv5 hybrids [46]. Similarly, system-level hybrids incorporate IoT and edge intelligence to compensate for on-device constraints [54], demonstrating that optimization extends beyond the neural architecture itself.

Several case studies demonstrate the successful deployment of lightweight hybrid architectures for face mask detection. For example, systems combining MobileNetV2 and VGG19 have been implemented for real-time surveillance, achieving high accuracy and robustness in drone-based monitoring and public space compliance checks. Other models, such as SSD with MobileNetV2, have been used for automated face mask compliance monitoring, highlighting the flexibility and effectiveness of these architectures in diverse real-world scenarios [50,51,52,59].

Hybrid detectors based on MobileNetV2 + SSD [45], YOLOv3 variants [53], and multi-component ensembles [2] consistently outperform standalone models in deployment scenarios requiring person-level localization, mask-wearing classification and real-time operation. Additionally, smart-city systems integrating detection models with sensor networks, edge-computing nodes and cloud-based analytics highlight the broader potential of hybrid architectures in scalable public-health monitoring [54]. These examples reinforce that hybrid architecture serves as a bridge between computational feasibility and large-scale real-world applicability. Table 4 summarizes the main hybrid architecture employed in face-mask detection, highlighting their backbones, detection heads and reported strengths.

In summary, hybrid architecture provides a principled balance between speed, accuracy and deployment feasibility, making them the dominant and most practical design strategy for real-time face-mask detection in embedded and resource-constrained environments.

4. Comparative Performance Analysis

The architectures discussed in Section 3, conventional CNNs (e.g., VGG, ResNet, DenseNet), lightweight models (e.g., MobileNetV2-based designs), and hybrid/attention-enhanced YOLO variants, are typically evaluated along two axes: recognition quality (accuracy-oriented metrics) and operational efficiency (latency, throughput, and resource usage). This section synthesizes reported results from recent face-mask detection studies to compare these families more explicitly.

4.1. Evaluation Metrics

Performance evaluation of face-mask detection systems relies on well-established metrics from classification and object-detection literature. These metrics quantify correctness, robustness, and discriminative capability of the models under various conditions (binary vs. multi-class classification, detection vs. recognition, occlusion, class imbalance, etc.). The primary metrics include Accuracy, Precision, Recall (Sensitivity) and F1-Score.

Let the outcomes for a mask-classification or detection system be described using the standard confusion-matrix counts:

• TP (True Positive): correctly predicted positive instances
• FP (False Positive): incorrect positive predictions
• FN (False Negative): missed positive instances
• TN (True Negative): correctly predicted negative instances

Where positive may refer to “mask,” “no-mask,” or “improper mask,” depending on the class definition.

4.1.1. Accuracy

Accuracy measures the proportion of correctly predicted instances among all predictions.

$$\mathrm{Accuracy}={\displaystyle \frac{TP+TN}{TP+TN+FP+FN}},$$

(1)

It is widely used in early face-mask classification (e.g., mask vs. no-mask), especially on curated datasets such as MaskedFace-Net. However, accuracy can be misleading in imbalanced datasets where one class dominates (e.g., many more masked than improperly masked samples). Therefore, accuracy should always be interpreted along with precision and recall.

4.1.2. Precision

Precision indicates the proportion of correct positive predictions out of all predicted positives.

$$\mathrm{Precision}={\displaystyle \frac{TP}{TP+FP}},$$

(2)

High precision means the model makes few false alarms (e.g., rarely labels someone as “no mask” incorrectly). For multi-class mask compliance (proper / improper / no mask), precision is computed per class, then averaged macro- or weighted-wise.

4.1.3. Recall (Sensitivity)

Recall represents the proportion of actual positive instances that are correctly detected.

$$\mathrm{Recall}={\displaystyle \frac{TP}{TP+FN}},$$

(3)

High recall means the model makes few misses, which is important for detecting improperly worn masks or no-mask cases in high-risk zones.

For mask detection systems deployed in public spaces, recall is often considered more critical than precision, since missing non-compliant subjects can compromise public safety.

4.1.4. F1-Score

The F1-score is the harmonic mean of precision and recall:

$$\mathrm{F}1\mathrm{-score}=2\times {\displaystyle \frac{\mathrm{Precision}\cdot \mathrm{Recall}}{\mathrm{Precision}+\mathrm{Recall}}}$$

(4)

It provides a balanced measure, particularly when the dataset is imbalanced. F1-score is widely reported in both conventional CNN mask classifiers and lightweight/hybrid models such as SSD-MobileNetV2, SE-YOLOv3, and YOLOv4-tiny derivatives.

Several deep-learning–based face mask detection studies [60,61,62,63] have reported accuracy, precision, recall, and F1-score as their primary evaluation metrics, typically derived from confusion matrices built over test sets or cross-validation folds. For example, Hussain et al. developed a MobileNetV2-based transfer learning system for binary mask classification and evaluated it on two datasets (one in-house, one public). They compared a generic DCNN with MobileNetV2 and reported per-model accuracy, precision, recall, and F1-score, showing that MobileNetV2 achieved up to 99% accuracy with closely matched precision, recall, and F1, thereby justifying model selection based on balanced performance across all four metrics rather than accuracy alone [60]. Similarly, Umer et al. proposed a custom CNN with a four-stage image-processing pipeline for face mask detection and evaluated it on their real-image RILFD dataset and two public datasets (MAFA and MOXA), comparing their model against YOLOv3 and Faster R-CNN; they reported detailed precision, recall, accuracy, and F1-scores across datasets, illustrating how these metrics jointly reveal robustness under real-world variations in lighting, mask type, and occlusion.

Ensemble and hybrid classification approaches also rely heavily on these four measures to demonstrate robustness under different data distributions. Bania et al. proposed an ensemble of ResNet50, Inception-v3, and VGG-16 for real-time mask detection and reported class-wise precision, recall, and F1, as well as macro-averaged scores; their best configuration achieved F1-scores around 0.997, indicating that both false positives and false negatives were extremely rare on their test sets [64]. Similarly, Habeeb et al. focused on incorrect facemask-wearing detection (e.g., mask under nose or chin) and explicitly emphasized that high recall was required to avoid missing non-compliant cases, reporting accuracy of 99.4%, precision of 99.4%, recall of 98.6%, and F1-score of 99.0% on their multi-class dataset [65]. These works highlight that precision and recall are often more informative than accuracy when the cost of missing violators is high.

Several architectures that jointly address face-mask detection and masked face recognition use the same four metrics to evaluate both tasks. Ullah et al. introduced DeepMaskNet, an end-to-end framework trained on their MDMFR dataset for detecting mask usage and recognizing identities under mask occlusion. They report accuracy, precision, recall, and F1-score for both detection and recognition modules, showing that the model maintains high recall and F1 even under variations in pose, illumination, mask type, and occlusion, thereby demonstrating that the chosen metrics are sensitive enough to capture performance degradation under real-world conditions [5]. More recent works on parallel or multi-branch CNNs for mask detection similarly present confusion matrices and derive all four metrics to compare variants with different backbones or feature-fusion strategies [48].

In object-detection–oriented approaches (e.g., YOLO or SSD variants), accuracy, precision, recall, and F1-score are frequently complemented by detection-specific metrics such as Average Precision (AP) and mean Average Precision (mAP), which jointly evaluate both localisation and classification performance. For instance, Kumar et al. employed these metrics in their ETL-YOLOv4 framework, showing improvements in mAP across multiple IoU thresholds when enhancing the YOLOv4 backbone for mask-related detection tasks. Beyond this, several recent studies have demonstrated how mAP, along with precision–recall curves, provides deeper insights into robustness under real-world variability [66]. Sheikh and Zafar evaluated their RRFMDS real-time detection system using accuracy, precision, recall, F1-score, and mAP, demonstrating that balanced behavior across these metrics is essential for consistent detection in crowded surveillance settings [15]. Similarly, Umer et al. compared their CNN-based detector against YOLOv3 and Faster R-CNN using accuracy, precision, recall, F1-score and AP values on multiple datasets, showing how detection-oriented metrics reveal performance gaps in scenarios involving occlusion, lighting variation, and mask-wearing diversity [62]. Collectively, these findings indicate that AP (Average Precision), mAP (Mean Average Precision) and precision-recall analysis play a crucial role when evaluating detection-centric mask-monitoring systems, especially for models deployed in unconstrained or real-time environments.

Survey and review papers on masked-face detection and recognition reinforce the central role of these four metrics. Hosny et al. systematically analyze deep-learning–based masked face detection methods and observe that almost all recent works report at least accuracy and F1-score, with many also giving precision and recall to capture trade-offs between false alarms and missed detections [67]. Further, several studies and surveys note that additional metrics such as specificity (true-negative rate), Receiver Operating Characteristic (ROC) curves, Area Under the ROC Curve (AUC), and Intersection over Union (IoU), are increasingly employed alongside accuracy, precision, recall, and F1-score to provide a more comprehensive picture of model performance, especially for real-time, safety-critical deployments and highly imbalanced datasets [1,6,7,8,9,68,69,70,71,72,73]. Table 5 summarizes which evaluation metrics are reported across a representative subset of face mask detection studies.

Notably, only a few recent works explicitly report ROC curves and AUC values for face-mask detection, such as ensemble-based detectors and IoT access-control systems, whereas most studies rely primarily on accuracy, precision, recall, F1-score, and mAP.

4.2. Trade-Offs Between Accuracy and Efficiency

The trade-off between accuracy and computational efficiency is a central challenge in the development and deployment of deep learning models for face-mask detection. Heavyweight architectures, such as VGGNet [18], ResNet [21], DenseNet [22], and Inception variants [19,20], typically achieve high accuracy on curated datasets but incur substantial computational and memory demands. Their high parameter counts, slow inference speeds on embedded processors, and significant GPU memory requirements make them less suitable for real-time surveillance or IoT applications with strict latency and power constraints. Consequently, practical deployments increasingly favor architectures that aim to balance representational capacity with computational efficiency, especially in resource-constrained public health monitoring scenarios.

4.2.1. Impact of Model Size and Architecture on Accuracy and Efficiency

Lightweight models, including MobileNet, EfficientNet, ShuffleNet, GhostNet, YOLOv3-tiny, and YOLOv4-tiny, reduce parameter count and inference latency by design, enabling fast deployment on embedded or mobile devices. For example, YOLOv4-tiny contains approximately six million parameters, nearly one-tenth of YOLOv4, and offers significantly higher detection speed, making it more suitable for real-time applications while still incurring a reduction in detection accuracy compared to the full model [74,75]. Similar observations are reported across lightweight design attempts: decreasing the dimensionality of feature maps can reduce inference time but often results in higher false detections. For instance, LSNet reduces feature dimensionality by 35% but increases false-detection rates by 41%, whereas models like SwinNet (198M parameters) achieve higher accuracy but remain computationally prohibitive for real-time edge deployment [37].

Some lightweight models nonetheless demonstrate strong balanced performance. MobileNetV2 frequently achieves accuracy around 92–93% with F1-scores above 0.90 in real-time scenarios [62,75]. Similarly, an ultra-lightweight model with only 0.12M parameters achieves competitive accuracy (95.41% for binary and 95.54% for three-class classification), showing that significant parameter reduction is possible without proportional loss in performance [76].

4.2.2. Comparative Performance of Lightweight and Heavyweight Models

Heavyweight models such as Mask R-CNN with FPN backbones, DenseNet201, or ConvNeXt-T variants achieve high accuracy (often approaching 99%) but are unsuitable for real-time edge deployment due to their size and slow inference speeds [77,78,79]. For example, Mask R-CNN ConvNeXt-T FPN provides state-of-the-art accuracy but is impractical for hardware-limited environments [80].

Hybrid and ensemble approaches offer a middle ground. Combining lightweight and heavyweight components can improve accuracy while preserving efficiency. An ensemble of single- and two-stage detectors achieved approximately 98.2% accuracy with an average inference time of 0.05 seconds, demonstrating that such hybrid strategies can meet both accuracy and latency requirements in real-time deployments [2]. Many studies emphasize that model selection for deployment should reflect hardware limitations and operational scenarios, balancing accuracy with speed, power consumption, and available compute resources [74,75,80,81].

To better contextualize these performance trends, it is necessary to compare the underlying architectural families that shape each model’s computational behavior. Differences in parameter count, convolutional design, memory footprint, and inference throughput directly determine whether architecture can meet real-time requirements on edge devices or embedded systems. A structural comparison of the major architecture families used in face-mask detection is presented in Table 6, highlighting differences in parameter size, inference speed, memory consumption, and suitability for edge deployment.

This contextual foundation supports the subsequent analysis of accuracy–efficiency patterns across representative models. The strategies discussed next build upon these architectural characteristics to further enhance model performance under resource constraints. By understanding the inherent computational profiles of each architecture family, it becomes clearer how transfer learning, data augmentation, and model-level modifications can be applied to strengthen accuracy without compromising real-time efficiency.

Strategies to Improve the Trade-Off

Transfer Learning: Utilizing pre-trained models and transfer learning can enhance the accuracy of lightweight models without significantly increasing computational demands. For example, transfer learning with MobileNetV2 and ResNet50 has been shown to improve detection accuracy in real-time applications [2,9,62].

Data Augmentation and Balanced Datasets: Techniques such as random over-sampling and data augmentation help improve model accuracy by addressing class imbalance, as seen in studies that reduced imbalance ratios and achieved high accuracy with efficient models [2,76].

Model Modifications: Modifying backbone networks, activation functions, and loss functions, as done in YOLOv4 and its variants, can help maintain good accuracy while improving speed and reducing resource consumption [9,74].

The comparative relationships between accuracy and computational efficiency across representative models are summarized in Table 7 providing a consolidated view of the trade-offs discussed in this section.

Overall, while conventional CNNs deliver high accuracy, their computational cost limits deployment on real-time, hardware-constrained platforms. Lightweight architectures offer high efficiency but may lack robustness in complex scenes. Hybrid and ensemble frameworks consistently offer the most practical balance between accuracy and speed, making them suitable for real-time monitoring in smart-city, IoT, and embedded environments. Thus, the optimal choice depends on the deployment context: accuracy-critical applications may tolerate heavier models, whereas large-scale real-time monitoring benefits most from efficient lightweight or hybrid architectures.

5. Future Research Directions

Despite the significant progress achieved through conventional, lightweight, and hybrid architectures, several open challenges continue to limit the robustness, generalizability, and scalability of face-mask detection systems. These challenges present rich opportunities for future research, particularly in improving multi-class capability, domain adaptation, model compression, and extending mask-detection frameworks toward broader applications within public-health monitoring and intelligent surveillance.

5.1. Improper Mask Detection and Multi-Class Analysis

Most existing models are optimized for binary mask detection (mask vs. no mask), yet improper mask wearing remains a major real-world compliance issue. Identifying masks worn below the nose, under the chin, partially covering the mouth, or loosely attached requires fine-grained, region-aware feature extraction that many lightweight systems struggle with. Conventional CNNs offer strong discriminative capacity but are computationally expensive, while lightweight architectures may misclassify subtle misuse cases due to reduced representational depth. Hybrid detectors improve robustness, but their performance still degrades when improper mask patterns exhibit high variability across individuals, mask materials, or face shapes.

Future work must therefore address multi-class imbalance, generate or augment datasets with diverse improper-wear scenarios, and incorporate region-specific constraints or anatomical priors. Approaches such as fine-grained visual reasoning, dense part-based attention, or landmark-guided mask positioning analysis could substantially improve improper-mask detection accuracy.

5.2. Domain Adaptation and Real-World Variability

A persistent challenge across all model families is domain shift, the discrepancy between curated training datasets and real-world deployment conditions. Surveillance environments introduce uncontrolled variables including illumination changes, shadows, motion blur, camera angle variations, diverse mask designs, facial accessories, occlusions (hands, hair, objects), and crowd density. Models trained purely on curated datasets such as MaskedFace-Net or PWMFD often suffer significant accuracy drops when deployed in unseen domains, demonstrating insufficient generalization.

Future research must explore domain-adaptation strategies, such as:

• Unsupervised domain adaptation (UDA) for aligning feature distributions across environments.
• Self-supervised representation learning to reduce dependency on labels
• Cross-dataset training pipelines that incorporate heterogeneous noise, mask materials, and cultural variations
• Synthetic domain randomization to simulate low-quality or occluded footage

Enhancing robustness under real-world variability is critical for scalable deployment across transportation hubs, hospitals, campuses, and crowded public spaces.

5.3. Expanding Applications Beyond Mask Detection

As mask usage declines post-pandemic, face-mask detection systems should evolve toward broader public-safety, healthcare, and compliance-monitoring applications. Since many architectural foundations are adaptable such as lightweight CNNs, attention-enhanced detectors, and hybrid YOLO variants, researchers can repurpose these systems for:

• PPE compliance monitoring (helmets, gloves, lab coats)
• Human behavior analysis (face-touching detection, cough detection, proximity violations)
• Health screening (visible respiratory cues, temperature screening integration)
• Access-control and identity verification under occlusion
• Crowd analytics and anomaly detection for smart-city infrastructure

Furthermore, integrating mask detection with multimodal sensing (audio, thermal imaging, RFID) can enable holistic public-health monitoring systems. The shift from task-specific detectors to generalizable compliance-monitoring frameworks represents a major future research opportunity.

6. Conclusion

This review analyzed the architectural evolution of deep learning–based face mask detection systems, tracing the progression from conventional CNNs to lightweight and hybrid architectures designed for real-time deployment on resource-constrained platforms. While traditional models such as VGGNet, ResNet, and DenseNet offer strong feature extraction capabilities, their high computational and memory demands limit their practicality for embedded and IoT applications. Lightweight architectures, including MobileNet, EfficientNet, ShuffleNet, and other efficient convolutional families, significantly reduce inference latency and parameter count, enabling practical deployment without severe accuracy degradation. Hybrid models, which integrate lightweight backbones with optimized detection heads or multi-branch processing strategies, further advance the accuracy–efficiency balance and represent the most promising direction for scalable real-time compliance monitoring.

The comparative performance analysis demonstrated that no single architecture simultaneously maximizes accuracy, robustness, and efficiency; rather, each model family offers strengths tuned to particular deployment requirements. Open challenges remain in improper-mask detection, domain adaptation to real-world variability, and the need for compression-oriented techniques such as pruning and knowledge distillation. Future research must therefore focus on developing architectures that generalize robustly across environments, operate reliably under multi-class conditions, and support energy-efficient computation on embedded devices. Additionally, as mask usage transitions from pandemic-driven contexts to broader occupational safety applications, the underlying architectural principles reviewed here provide a foundation for more general compliance-monitoring systems.

Overall, this review contributes an architecture-centered perspective on face mask detection, clarifying design principles, identifying limitations, and outlining research directions to support the development of efficient, reliable, and scalable monitoring systems suitable for modern edge-computing environments