Teacher Network Research Articles

Information gap based knowledge distillation for occluded facial expression recognition

Facial Expression Recognition (FER) with occlusion presents a challenging task in computer vision because facial occlusions result in poor visual data features. Recently, the region attention technique has been introduced to address this problem by researchers, which make the model perceive occluded regions of the face and prioritize the most discriminative non-occluded regions. However, in real-world scenarios, facial images are influenced by various factors, including hair, masks and sunglasses, making it difficult to extract high-quality features from these occluded facial images. This inevitably limits the effectiveness of attention mechanisms. In this paper, we observe a correlation in facial emotion features from the same image, both with and without occlusion. This correlation contributes to addressing the issue of facial occlusions. To this end, we propose a Information Gap based Knowledge Distillation (IGKD) to explore the latent relationship. Specifically, our approach involves feeding non-occluded and masked images into separate teacher and student networks. Due to the incomplete emotion information in the masked images, there exists an information gap between the teacher and student networks. During training, we aim to minimize this gap to enable the student network to learn this relationship. To enhance the teacher’s guidance, we introduce a joint learning strategy where the teacher conducts knowledge distillation on the student during the training of the teacher. Additionally, we introduce two novel constraints, called knowledge learn and knowledge feedback loss, to supervise and optimize both the teacher and student networks. The reported experimental results show that IGKD outperforms other algorithms on four benchmark datasets. Specifically, our IGKD achieves 87.57% on Occlusion-RAF-DB, 87.33% on Occlusion-FERPlus, 64.86% on Occlusion-AffectNet, and 73.25% on FED-RO, clearly demonstrating its effectiveness and robustness.

Fast-Vid2Vid++: Spatial-Temporal Distillation for Real-Time Video-to-Video Synthesis.

Video-to-Video synthesis (Vid2Vid) gains remarkable performance in generating a photo-realistic video from a sequence of semantic maps, such as segmentation, sketch and pose. However, this pipeline is heavily limited to high computational cost and long inference latency, mainly attributed to two essential factors: 1) network architecture parameters, 2) sequential data stream. Recently, the parameters of image-based generative models have been significantly reduced via more efficient network architectures. Existing methods mainly focus on slimming network architectures but ignore the size of the sequential data stream. Moreover, due to the lack of temporal coherence, image-based compression is not sufficient for the compression of the video task. In this paper, we present a spatial-temporal hybrid distillation compression framework, Fast-Vid2Vid++, which focuses on knowledge distillation of the teacher network and the data stream of generative models on both space and time. Fast-Vid2Vid++ makes the first attempt at time dimension to transfer hierarchical features and time coherence knowledge to reduce computational resources and accelerate inference. Specifically, we compress the data stream spatially and reduce the temporal redundancy. We distill the knowledge of the hierarchical features and the final response from the teacher network to the student network in high-resolution and full-time domains. We transfer the long-term dependencies of the features and video frames to the student model. After the proposed spatial-temporal hybrid knowledge distillation (Spatial-Temporal-HKD), our model can synthesize high-resolution key-frames using the low-resolution data stream. Finally, Fast-Vid2Vid++ interpolates intermediate frames by motion compensation with slight latency and generates full-length sequences with motion-aware inference (MAI). On standard benchmarks, Fast-Vid2Vid++ achieves a real-time performance of 30-59 FPS and saves 28-35× computational cost on a single V100 GPU.

A fast differential network with adaptive reference sample for gaze estimation

Most non-invasive gaze estimation methods do not consider the inter-individual differences in anatomical structure, but directly regress the gaze direction from the appearance image information, which limits the accuracy of individual-independent gaze estimation networks. In addition, existing gaze estimation methods tend to consider only how to improve the model’s generalization performance, ignoring the crucial issue of efficiency, which leads to bulky models that are difficult to deploy and have questionable cost-effectiveness in practical use. This paper makes the following contributions: (1) A differential network for gaze estimation using adaptive reference samples is proposed, which can adaptively select reference samples based on scene and individual characteristics. (2) The knowledge distillation is used to transfer the knowledge structure of robust teacher networks into lightweight networks so that our networks can execute quickly and at low computational cost, dramatically increasing the prospect and value of applying gaze estimation. (3) Integrating the above innovations, a novel fast differential neural network (Diff-Net) named FDAR-Net is constructed and achieved excellent results on MPIIGaze, UTMultiview and EyeDiap.

Semi-PKD: Semi-supervised Pseudoknowledge Distillation for saliency prediction

In saliency prediction, Knowledge Distillation (KD) is leveraged to improve the predictive performance of compact Student Networks. However, the challenge is searching for an optimal teacher–student pair while handling the unavailability of large-scale annotations in the Pseudoknowledge Distillation (PKD). To overcome this challenge, a semi-supervised method is proposed; Semi-PKD. This method involves pseudo-label generation on unlabeled data by a Teacher Network trained using the exponential moving average KD (EMA-KD) method. The EMA-KD method utilizes only the Student Network by acquiring self-knowledge, solving the problem of optimal teacher–student pair selection. Semi-PKD outperforms other state-of-the-art saliency prediction models across various evaluation metrics. The code is available at https://github.com/chakkritte/Semi-PKD.

Calibrated Adaptive Teacher for Domain-Adaptive Intelligent Fault Diagnosis

Intelligent fault diagnosis (IFD) based on deep learning can achieve high accuracy from raw condition monitoring signals. However, models usually perform well on the training distribution only, and experience severe performance drops when applied to a different distribution. This is also observed in fault diagnosis, where assets are often operated in working conditions different from the ones in which the labeled data have been collected. The scenario where labeled data are available in a source domain and only unlabeled data are available in a target domain has been addressed recently by unsupervised domain adaptation (UDA) approaches for IFD. Recent methods have relied on self-training with confident pseudo-labels for the unlabeled target samples. However, the confidence-based selection of pseudo-labels is hindered by poorly calibrated uncertainty estimates in the target domain, primarily due to over-confident predictions, which limits the quality of pseudo-labels and leads to error accumulation. In this paper, we propose a novel method called Calibrated Adaptive Teacher (CAT), where we propose to calibrate the predictions of the teacher network on target samples throughout the self-training process, leveraging post hoc calibration techniques. We evaluate CAT on domain-adaptive IFD and perform extensive experiments on the Paderborn University (PU) benchmark for fault diagnosis of rolling bearings under varying operating conditions, using both time- and frequency-domain inputs. We compare four different calibration techniques within our framework, where temperature scaling is both the most effective and lightweight one. The resulting method—CAT+TempScaling—achieves state-of-the-art performance on most transfer tasks, with on average 7.5% higher accuracy and 4 times lower calibration error compared to domain-adversarial neural networks (DANNs) across the twelve PU transfer tasks.

A Network Analysis of Emotional Intelligence in Chinese Preschool Teachers

Emotional intelligence significantly influences various aspects of teachers’ professional and personal lives, notably affecting preschoolers’ social skills and emotional development during formative years. This study utilizes a network analysis approach to explore the complex relationships among four components of emotional intelligence: emotional labor, emotional regulation, professional well-being, and professional identity. Participants included 2069 frontline Chinese teachers (34 males, 2035 females; M = 28.64, SD = 5.98; M years of teaching = 6.88, SD = 5.29) with no leadership roles, categorized into three stages of their careers based on years of teaching experience: novice (0–3 years; n = 612), advanced beginners (4–6 years; n = 537), and competent (7–40 years; n = 920). Findings revealed that joy of teaching, role value, and professional value were identified as the most critical elements within the emotional state network of early childhood education teachers. The strongest connections in teachers’ emotional networks were found between school connectedness and joy of teaching (r = 0.474), surface acting behavior and natural acting behavior (r = 0.419), and professional value and professional behavior (r = 0.372). Furthermore, teachers across different career stages exhibited similar characteristics and intrinsic connections among emotional state components. These findings deepen our understanding of the emotional state networks of ECE teachers, highlighting shared features and interconnected mechanisms, and suggest that enhancing teachers’ emotional intelligence through targeted professional development can improve both teacher well-being and preschoolers’ social–emotional outcomes. Policies that foster strong school connectedness and reduce emotional labor are key to promoting sustained joy in teaching, particularly for novice and advanced beginner teachers.

Simplified Knowledge Distillation for Deep Neural Networks Bridging the Performance Gap with a Novel Teacher–Student Architecture

The rapid evolution of deep learning has led to significant achievements in computer vision, primarily driven by complex convolutional neural networks (CNNs). However, the increasing depth and parameter count of these networks often result in overfitting and elevated computational demands. Knowledge distillation (KD) has emerged as a promising technique to address these issues by transferring knowledge from a large, well-trained teacher model to a more compact student model. This paper introduces a novel knowledge distillation method that simplifies the distillation process and narrows the performance gap between teacher and student models without relying on intricate knowledge representations. Our approach leverages a unique teacher network architecture designed to enhance the efficiency and effectiveness of knowledge transfer. Additionally, we introduce a streamlined teacher network architecture that transfers knowledge effectively through a simplified distillation process, enabling the student model to achieve high accuracy with reduced computational demands. Comprehensive experiments conducted on the CIFAR-10 dataset demonstrate that our proposed model achieves superior performance compared to traditional KD methods and established architectures such as ResNet and VGG networks. The proposed method not only maintains high accuracy but also significantly reduces training and validation losses. Key findings highlight the optimal hyperparameter settings (temperature T = 15.0 and smoothing factor α = 0.7), which yield the highest validation accuracy and lowest loss values. This research contributes to the theoretical and practical advancements in knowledge distillation, providing a robust framework for future applications and research in neural network compression and optimization. The simplicity and efficiency of our approach pave the way for more accessible and scalable solutions in deep learning model deployment.

MonoFG: Monocular 3D Object Detection with Knowledge Distillation for Human-Centric Autonomous Driving Systems

Monocular 3D object detection is essential for identifying objects in road images, thus offering valuable environmental perception data that are crucial for human-centric autonomous driving systems. However, due to the inherent limitations of camera imaging, obtaining precise depth information from images alone is challenging, which hampers the accuracy of within-scene object localization. In this paper, we introduce a monocular 3D object detection method called MonoFG that uses knowledge distillation with separated foreground and background components to improve the accuracy of object localization. First, detached foreground and background distillation processes can strategically leverage the distinct positional information acquired from each location to optimize the produced global distillation effects. This step serves as the foundation for the subsequent feature and response distillation process, which focuses on the distilled foreground and background rather than isolated object distillation. Second, triple attention mechanism-based feature distillation intensifies the feature imitation and feature representation capabilities of the student network. Spatial and channel attention mechanisms encourage the student network to capture crucial pixels and channels from the teacher network, whereas a self-attention mechanism globally transfers the learned relationships between pixels. Third, localization error-based response distillation facilitates a clearer transfer of positional information from the teacher network to the student network. Only when the positioning ability of the teacher network exceeds that of the student network can knowledge be comprehensively distilled across both the foreground and background. Therefore, the distillation process is constrained to specific content, which is delineated by positioning errors that serve as the boundaries. Finally, experiments conducted on the KITTI benchmark dataset demonstrate that our method outperforms many well-known baseline methods in several representative evaluation tasks (e.g., 3D object detection and bird's-eye view (BEV) detection) involving human-centric autonomous driving systems.

Data free knowledge distillation with feature synthesis and spatial consistency for image analysis

Privacy and security concerns restrict access to original training datasets, posing significant challenges for model compression. Data-Free Knowledge Distillation (DFKD) emerges as a solution, aiming to transfer knowledge from teacher to student networks without accessing original data. Existing DFKD methods struggle to generate high-quality synthetic samples that capture the complexities of real-world data, leading to suboptimal knowledge transfer. Moreover, these approaches often fail to preserve the spatial attributes of the teacher network, resulting in shortcut learning and limited generalization.To address these issues, a novel DFKD strategy is proposed with three innovations: (1) an enhanced DCGAN generator with an attention module for synthesizing samples with improved micro-discriminative features; (2) a Multi-Scale Spatial Activation Region Consistency (MSARC) mechanism to accurately replicate the teacher’s spatial attributes; and (3) an adversarial learning framework that creates a dynamic competitive environment between the generative and distillation phases. Rigorous evaluation of the method on several benchmark datasets, including CIFAR-10, CIFAR-100, Tiny-ImageNet, and medical imaging datasets such as PathMNIST, BloodMNIST, and PneumoniaMNIST, demonstrates superior performance compared to existing DFKD methods. Specifically, on CIFAR-100, the student network attains an accuracy of 77.85%, surpassing previous methods like CMI and SpaceshipNet. On BloodMNIST, the method achieves an accuracy of 80.50%, outperforming the next best method by over 5%.

Learning face super-resolution through identity features and distilling facial prior knowledge

Deep learning techniques in electronic surveillance have shown impressive performance for super-resolution (SR) of captured low-quality face images. Most of these techniques adopt facial priors to improve the feature details in the resultant super-resolved images. However, the estimation of facial priors from the captured low-quality images is often inaccurate in real-life situations because of their tiny, noisy, and blurry nature. Thus, the fusion of such priors badly affects the performance of these models. Therefore, this work presents a teacher–student-based face SR framework that efficiently preserves the personal facial structure information in the super-resolved faces. In the proposed framework, the teacher network exploits the facial heatmap-based ground-truth-prior to learn the facial structure that is utilized by the student network. The student network is trained with the identity feature loss for maintaining the identity and facial structure information in reconstructed high-resolution (HR) face images. The performance of the proposed framework is evaluated by conducting the experimental study on standard datasets namely CelebA-HQ and LFW face. The experimental results reveal that the proposed technique conquers the existing methods for the face SR task.

Multidimensional knowledge distillation for multimodal scene classification of remote sensing images

The advancement of deep learning technology has significantly improved the performance of remote sensing image (RSI) scene classification. However, it is important to note that most RSI scene classification models heavily depend on complex structures, resulting in high computational requirements and substantial costs. This study addresses this issue by utilizing a state-of-the-art model compression technique known as knowledge distillation (KD). The objective of KD is to transfer extensive knowledge from an excellent teacher model to a lightweight student model. While existing models focus on guiding the student network to learn specific stage or scale features from the teacher network, they lack comprehensiveness. To enhance the model's feature representation capability in complex scenarios, this study proposes a multidimensional KD approach (MKD). MKD enables the student network (MKD-S) to learn the feature representation capability of the teacher network (MKD-T) at each stage through a hybrid KD method. Specifically, the encoder incorporates a local-global KD mechanism to capture both low-level local information and high-level global information based on feature differences. Moreover, the fusion stage introduces inter-layer relationship KD and intra-layer feature KD to account for the dependencies between intermediate features within the MKD-S and MKD-T models. Additionally, the discrete wavelet transform, known for its ability to capture frequency domain and time domain features, is applied in the decoding stage of the MKD-T. This integration of decoding features across layers enables the completion of the knowledge response in the MKD-S. Experimental results demonstrate the effectiveness of our MKD on two benchmark datasets: Vaihingen and Potsdam.

PE-MCAT: Leveraging Image Sensor Fusion and Adaptive Thresholds for Semi-Supervised 3D Object Detection.

Existing 3D object detection frameworks in sensor-based applications heavily rely on large-scale annotated data to achieve optimal performance. However, obtaining such annotations from sensor data-like LiDAR or image sensors-is both time-consuming and costly. Semi-supervised learning offers an efficient solution to this challenge and holds significant potential for sensor-driven artificial intelligence (AI) applications. While it reduces the need for labeled data, semi-supervised learning still depends on a small amount of labeled samples for training. In the initial stages, relying on such limited samples can adversely affect the effective training of student-teacher networks. In this paper, we propose PE-MCAT, a semi-supervised 3D object detection method that generates high-precision pseudo-labels. First, to address the challenges of insufficient local feature capture and poor robustness in point cloud data, we introduce a point enrichment module. This module incorporates information from image sensors and combines multiple feature fusion methods of local and self-features to directly enhance the quality of point clouds and pseudo-labels, compensating for the limitations posed by using only a few labeled samples. Second, we explore the relationship between the teacher network and the pseudo-labels it generates. We propose a multi-class adaptive threshold strategy to initially filter and create a high-quality pseudo-label set. Furthermore, a joint variable threshold strategy is introduced to refine this set further, enhancing the selection of superior pseudo-labels.Extensive experiments demonstrate that PE-MCAT consistently outperforms recent state-of-the-art methods across different datasets. Specifically, on the KITTI dataset and using only 2% of labeled samples, our method improved the mean Average Precision (mAP) by 0.7% for cars, 3.7% for pedestrians, and 3.0% for cyclists.

Effective Bi-decoding networks for rail-surface defect detection by knowledge distillation

No-service rail-surface defect detection is a crucial method for assessing the quality of railroad tracks. However, the low-contrast and dark-tone characteristics of track-surface textures pose challenges to current defect-monitoring techniques. Real-time and on-site online inspections are important to ensure safe railway operation; however, most complex models for no-service inspections are difficult to deploy on mobile devices. To address these challenges and overcome the detection difficulties associated with complex scenes, we designed a knowledge distillation-based double decoding-layer refinement network (EBDNet-KD). The first decoding process is guided by a bimodal high-level semantic feature map obtained by extending the attention-based graph convolution to incrementally enhance the dual-stream features and obtain an image restoration prior. A divide-and-conquer decoder is then designed to distinguish features using different decoding layers. The prior is then used in the second decoding layer, which enables the bimodal features to interact fully and obtain the final prediction map. We introduce a knowledge distillation strategy that enables a lightweight, compact student network to learn a complex teacher network’s feature extraction process. This facilitates pixel-consistent learning of the knowledge within the bi-decoder layer, as well as bidirectional learning of the focused contextual response knowledge to optimize the model. The EBDNet-KD significantly reduces computational costs while guaranteeing performance with a parameter count of only 28 M. EBDNet-KD demonstrated superior performance over 15 state-of-the-art methods in experiments conducted on NEU RSDDS-AUG, an industrial RGB-depth dataset. We assessed the generalizability of EBDNet-KD by evaluating its performance on three additional public datasets, yielding competitive results. The source code and results can be found at https://github.com/Wuyue15/EBDNet.

MT-Net: Single image dehazing based on meta learning, knowledge transfer and contrastive learning

Single image dehazing is becoming increasingly important as its results impact the efficiency of subsequent computer vision tasks. While many methods have been proposed to address this challenge, existing dehazing approaches often exhibit limited adaptability to different types of images and lack future learnability. In light of this, we propose a dehazing network based on meta-learning, knowledge transfer, and contrastive learning, abbreviated as MT-Net. In our approach, we combine knowledge transfer with meta-learning to tackle these challenges, thus enhancing the network’s generalization performance. We refine the structure of knowledge transfer by introducing a two-phases approach to facilitate learning under the guidance of teacher networks and learning committee networks. We also optimize the negative examples of contrastive learning to reduce the contrast space. Extensive experiments conducted on synthetic and real datasets demonstrate the remarkable performance of our method in both quantitative and qualitative comparisons. The code has been released on https://github.com/71717171fan/MT-Net.

Probabilistic 3D Correspondence Prediction from Sparse Unsegmented Images.

The study of physiology demonstrates that the form (shape) of anatomical structures dictates their functions, and analyzing the form of anatomies plays a crucial role in clinical research. Statistical shape modeling (SSM) is a widely used tool for quantitative analysis of forms of anatomies, aiding in characterizing and identifying differences within a population of subjects. Despite its utility, the conventional SSM construction pipeline is often complex and time-consuming. Additionally, reliance on linearity assumptions further limits the model from capturing clinically relevant variations. Recent advancements in deep learning solutions enable the direct inference of SSM from unsegmented medical images, streamlining the process and improving accessibility. However, the new methods of SSM from images do not adequately account for situations where the imaging data quality is poor or where only sparse information is available. Moreover, quantifying aleatoric uncertainty, which represents inherent data variability, is crucial in deploying deep learning for clinical tasks to ensure reliable model predictions and robust decision-making, especially in challenging imaging conditions. Therefore, we propose SPI-CorrNet, a unified model that predicts 3D correspondences from sparse imaging data. It leverages a teacher network to regularize feature learning and quantifies data-dependent aleatoric uncertainty by adapting the network to predict intrinsic input variances. Experiments on the LGE MRI left atrium dataset and Abdomen CT-1K liver datasets demonstrate that our technique enhances the accuracy and robustness of sparse image-driven SSM.

PPDistiller: Weakly-supervised 3D point cloud semantic segmentation framework via point-to-pixel distillation

Despite the significant growth in the availability of 3D light detection and ranging (LiDAR) point cloud data in recent years, annotation remains expensive and time-consuming. This has led to an increasing demand for weakly-supervised semantic segmentation (WSSS) methods in applications such as autonomous driving, mapping, and robotics. Existing approaches typically rely solely on LiDAR point cloud data for WSSS, which often results in lower segmentation accuracy due to the sparsity of point clouds. To address these challenges, we propose a novel architecture, PPDistiller, which employs multiple teacher networks from different modalities. Compared to other WSSS and multimodal approaches, PPDistiller achieves superior segmentation accuracy with fewer annotations. This is facilitated through the novel Mean Multi-Teacher Framework (MMT), which incorporates multiple modalities and teachers. To address the issue of lacking 2D labels, we propose the Distance-CAM Self-Training (DCAM-ST) module, which utilizes sparse 3D weak annotations to produce accurate 2D pixel-level annotations. To enable adaptive fusion of 2D and 3D data, we introduce the Attention Point to Pixel Fusion (APPF) module, facilitating bidirectional transfer of cross-modal knowledge. Additionally, to fully leverage the spatial semantic information in point cloud, we propose the Pyramid Semantic-context Neighbor Aggregation (PSNA) module, aiming to exploit spatial and semantic correlations to improve performance. Extensive experimentation on SemanticKITTI, ScribbleKITTI and nuScenes datasets demonstrates the superiority of our proposed method. Compared to state-of-the-art fusion and weakly-supervised methods, PPDistiller achieves the highest mean Intersection over Union (mIoU) scores under both fully-supervised and weakly-supervised settings.

Driver Distraction Detection Based on Fusion Enhancement and Global Saliency Optimization

Driver distraction detection not only effectively prevents traffic accidents but also promotes the development of intelligent transportation systems. In recent years, thanks to the powerful feature learning capabilities of deep learning algorithms, driver distraction detection methods based on deep learning have increased significantly. However, for resource-constrained onboard devices, real-time lightweight models are crucial. Most existing methods tend to focus solely on lightweight model design, neglecting the loss in detection performance for small targets. To achieve a balance between detection accuracy and network lightweighting, this paper proposes a driver distraction detection method that combines enhancement and global saliency optimization. The method mainly consists of three modules: context fusion enhancement module (CFEM), channel optimization feedback module (COFM), and channel saliency distillation module (CSDM). In the CFEM module, one-dimensional convolution is used to capture information between distant pixels, and an injection mechanism is adopted to further integrate high-level semantic information with low-level detail information, enhancing feature fusion capabilities. The COFM module incorporates a feedback mechanism to consider the impact of inter-layer and intra-layer channel relationships on model compression performance, achieving joint pruning of global channels. The CSDM module guides the student network to learn the salient feature information from the teacher network, effectively balancing the model’s real-time performance and accuracy. Experimental results show that this method outperforms the state-of-the-art methods in driver distraction detection tasks, demonstrating good performance and potential application prospects.

Domain generalization-based damage detection of composite structures powered by structural digital twin

This research addresses the challenge of generalizing deep learning models for different CFRP composite structures in the task of fatigue damage detection. To overcome this challenge, knowledge distillation is employed to enhance the generalizability of deep learning models. A teacher network processes continuous wavelet transform images using Fourier transform and neural networks, while a student network distills the teacher network. This framework improves the models' generalization performance by transferring knowledge from the teacher network to the student network. Additionally, soft gradient boosting is utilized to further enhance the generalizability. By constructing a main sub-network and multiple parallel auxiliary sub-networks within the teacher network, the student network mimics the main sub-network to achieve improved accuracy in the target domain and prevent overfitting. To augment limited datasets of real CFRP monitoring signals and help to learn domain-invariant features, structural digital twin technology is leveraged to generate simulated monitoring signals, which enables the models to capture domain invariant information, significantly enhancing its performance of fatigue damage detection across different structures. Damage detection based on the generalization results between multiple Layups demonstrates a test accuracy exceeding 80 % when the monitoring data of the target CFRP structure is unavailable during training. Therefore, the cross-structure damage detection ability of the proposed approach is well proved.

DOL3: Distilled OpenL3 audio embeddings for lightweight audio classification

Deep audio representations, also known as embeddings, recently became a popular alternative to conventional features like spectrograms for a wide range of audio classification tasks because of their domain-agnostic character and reduced training costs. Still, the usage is often limited to rather computationally intensive system due to the nature of their extraction from large networks. This paper aims to minimize the computational costs of embedding extraction by distilling the knowledge of the OpenL3 audio network to a smaller student network. Results show that the student network maintains comparable performance as the teacher network on various music and ambient noise classification tasks, while reducing the network size by over 90\% and the computational load by five times.

Memoryless Multimodal Anomaly Detection via Student–Teacher Network and Signed Distance Learning

Memoryless Multimodal Anomaly Detection via Student–Teacher Network and Signed Distance Learning