Spatio-temporal Representation Research Articles

Integrating Spatiotemporal and Travel-Related Information for Accurate Urban Passenger Profiling Using GANs

The elaborate description of passenger travel profiles is of significant importance in urban planning, socioeconomic structural design, and individual travel preference analysis. Traditional models often lack consideration of personalized features and exhibit suboptimal performance in constructing spatiotemporal dependencies. To address these issues, this paper proposes a method that integrates spatiotemporal information with travel-related information and employs generative adversarial networks (GANs) for adversarial training. This method accurately fits the true distribution of user travel data, thereby providing detailed profiles of public transportation passengers’ travel behavior. Specifically, the proposed approach considers the complete travel chain of individuals, establishes a spatiotemporal constraint representation model, and utilizes GANs to simulate the distribution of passenger travel, obtaining more compact and high-level travel vector features. The empirical results demonstrate that the proposed method accurately captures passengers’ travel patterns in both the temporal and spatial dimensions, offering technical support for urban transportation planning.

Highly Efficient and Unsupervised Framework for Moving Object Detection in Satellite Videos.

Moving object detection in satellite videos (SVMOD) is a challenging task due to the extremely dim and small target characteristics. Current learning-based methods extract spatio-temporal information from multi-frame dense representation with labor-intensive manual labels to tackle SVMOD, which needs high annotation costs and contains tremendous computational redundancy due to the severe imbalance between foreground and background regions. In this paper, we propose a highly efficient unsupervised framework for SVMOD. Specifically, we propose a generic unsupervised framework for SVMOD, in which pseudo labels generated by a traditional method can evolve with the training process to promote detection performance. Furthermore, we propose a highly efficient and effective sparse convolutional anchor-free detection network by sampling the dense multi-frame image form into a sparse spatio-temporal point cloud representation and skipping the redundant computation on background regions. Coping these two designs, we can achieve both high efficiency (label and computation efficiency) and effectiveness. Extensive experiments demonstrate that our method can not only process 98.8 frames per second on 1024 ×1024 images but also achieve state-of-the-art performance.

Convolution spatial-temporal attention network for EEG emotion recognition

In recent years, emotion recognition using electroencephalogram (EEG) signals has garnered significant interest due to its non-invasive nature and high temporal resolution. We introduced a groundbreaking method that bypasses traditional manual feature engineering, emphasizing data preprocessing and leveraging the topological relationships between channels to transform EEG signals from two-dimensional time sequences into three-dimensional spatio-temporal representations. Maximizing the potential of deep learning, our approach provides a data-driven and robust method for identifying emotional states. Leveraging the synergy between convolutional neural network and attention mechanisms facilitated automatic feature extraction and dynamic learning of inter-channel dependencies. Our method showcased remarkable performance in emotion recognition tasks, confirming the effectiveness of our approach, achieving average accuracy of 98.62% for arousal and 98.47% for valence, surpassing previous state-of-the-art results of 95.76% and 95.15%. Furthermore, we conducted a series of pivotal experiments that broadened the scope of emotion recognition research, exploring further possibilities in the field of emotion recognition.

Multi-view Traffic Flow Prediction Model Based on Spatiotemporal Convolution

Predicting traffic flow has always been a significant task in intelligent transportation systems. Due to the substantial temporal and spatial dependencies of traffic flow sequences, accurately predicting traffic flow poses a considerable challenge. Many existing works primarily rely on recurrent neural networks, graph neural networks, and Transformer models to establish traffic flow prediction models. To better extract features and enhance efficiency, a traffic flow prediction model based on multi-view spatiotemporal convolution (MVSC) is proposed. This model learns the representation of sequence data at the input encoding layer and incorporates location and time information. In the spatiotemporal feature representation learning layer, considering the diverse periodic patterns in sequences, several representation learning modules are designed, conducting local spatiotemporal feature exploration through one-dimensional convolution and then accomplishing global spatiotemporal feature mining based on causal convolution. To further enhance the model's utilization of spatiotemporal features, a channel attention mechanism is introduced at the prediction layer. The forecasting method employed in the study is direct multistep, and subsequent experiments conducted on two real datasets demonstrate that the MVSC model exhibits a certain degree of superiority in MAE, RMSE, and MAPE for both short-term and long-term predictions compared to existing models. And through the latest experiments and investigations, it has been found that MVSC has improved MAPE performance by about 1.2% compared to recent models such as RTGCN and STRGCN, achieving the intended outcomes.

SceneKG: A Geo‐Scene Based Spatiotemporal Knowledge Representation Framework Considering Geo‐Processes

ABSTRACTCurrent geographic knowledge graph research mainly focuses on static knowledge coupling through analysis of web text and maps, but its knowledge expression model fails to incorporate knowledge mechanisms such as process evolution and interactions. The geo‐scene is a synthesis of various geographic elements which prescribes spatial, temporal, semantic, attribute, and interrelationships. It has demonstrated its utility in facilitating geocomputation, spatiotemporal simulation and reasoning related to geo‐process. Therefore, developing a geo‐scene‐based knowledge graph is significant to improve the ability of spatiotemporal knowledge representation and service. In this study, we first analyze the basic characteristics of geo‐scenes and try to emphasize the evolution of geo‐scenes through redefining the ontologies ‘event’, ‘process’ and ‘state’. To describe the dynamic nested structure of geo‐scenes, we propose a geo‐scene knowledge representation framework, and illustrate the representation of a geo‐scene and its inner relationships. Moreover, to keep the integrity, interaction and dynamic of geo‐scenes, we develop a event‐process centered geo‐scene knowledge organization method. Furthermore, we have tested the utility of our proposed representation by the Beijing 731 heavy rainstorm. The case study represents heavy rainstorm to demonstrate the complicated structure of the disaster scene, and the support for precipitation process simulation, early warning information organization, and disaster situation information query. The proposed representation provides an idea for the construction of geo‐dynamic knowledge graphs, encodes both static and dynamic geographic knowledge, makes explicit the evolution and interaction knowledge in geo‐scenes, and expands the application service of geo‐scene knowledge graph in the era of GeoAI.

DEBEcoMod: A dynamic energy budget R tool to predict life-history traits of marine organisms across time and space

DEBEcoMod is an open-source R script designed to apply Dynamic Energy Budget (DEB) theory to predict life-history traits of marine organisms under various environmental and anthropogenic stressors. It presents a novel approach to overcoming the computational and scale limitations of previous DEB applications, enabling the generation of spatially explicit outputs. DEBEcoMod is intended to predict traits such as maximum size, reproductive output, and life-history traits across different temporal and spatial scales. It utilises parameters from the AddMyPet database for various species and environmental time series to simulate the past, present, and future performance of organisms. The tool also includes a module for spatio-temporal representation, producing clear and accessible maps for stakeholders. The document highlights DEBEcoMod's application in invasion biology, marine spatial planning, integrated multi-trophic aquaculture, and marine ecology, drawing on published examples of spatial applications to demonstrate its versatility and potential in ecological research and adaptive management. Furthermore, the code has been cross-validated with the official DEBtool to ensure its accuracy and reliability. DEBEcoMod is available for download on GitHub, enhancing its accessibility and utility for a wide range of ecological and conservation applications.

Dynamic graph structure and spatio-temporal representations in wind power forecasting

Wind Power Forecasting (WPF) has gained considerable focus as a crucial aspect of the successful integration and operation of wind power. However, due to the stochastic and unstable nature of wind, it poses a real challenge to effectively analyze the correlations among multiple time series data for accurate prediction. In our study, an end-to-end framework called Dynamic Graph structure and Spatio-Temporal representation learning (DSTG) framework is proposed to achieve stable power forecasting by constructing graph data to capture the critical features in the data. Specifically, a graph structure learning (GSL) module is introduced to dynamically construct task-related correlation matrices via backpropagation to mitigate the inherent inconsistency and randomness of wind power data. Additionally, a dual-scale temporal graph learning (DTG) module is further proposed to explore the implicit spatio-temporal features at a fine-grained level using different skip connections from the constructed graph data. Finally, comprehensive experiments are performed on the collected Xuji Group Wind Power (XGWP) dataset, and the results show that DSTG outperforms the state-of-the-art spatio-temporal methods by 10.12% on the average of root mean square error and mean absolute error, demonstrating the effectiveness of DSTG. In conclusion, our model provides a promising approach.

Towards accurate anomaly detection for cloud system via graph-enhanced contrastive learning

As a critical technology, anomaly detection ensures the smooth operation of cloud systems while maintaining the market competitiveness of cloud service providers. However, the resource data in real-world cloud systems is predominantly unannotated, leading to insufficient supervised signals for anomaly detection. Moreover, complicated topological associations existed between cloud servers (e.g., computation, storage, and communication). While acquiring resource information, correlating the system topology is challenging. To this end, we propose the GCAD for cloud system anomaly detection, which integrates data augmentation, GraphGRU, contrastive learning, and reconstruction. First, GCAD constructs positive and negative sample pairs through the masking and Gaussian noise data augmentation. Then, the GraphGRU processes extended temporal graph data, extracting and fusing spatiotemporal features from resource status and system topology. In addition, GCAD introduces linear attention for encoding spatiotemporal representations to capture their global correlation information. The weight parameters of the encoder are optimized using a contrastive learning mechanism. Finally, GCAD utilizes a reconstruction technique to calculate anomaly scores, facilitating the evaluation of the state of the cloud system at each time point. Experimental results indicate that GCAD outperforms state-of-the-art compared methods on two real-world datasets that contain topology information.

Baseline-free assisted lamb wave-based damage detection in CFRP composites using graph convolutional networks and Transformer models

Carbon fiber-reinforced polymer (CFRP) composites are widely used in aerospace but are susceptible to damage, threatening structural safety. Thus, developing an effective structural health monitoring system is essential. This study presents CFRP-former, a novel deep learning framework that detects damage in CFRP composites by constructing spatio-temporal representation graphs of Lamb waves (LWs). The CFRP-former comprises three key modules: graph convolutional neural network (GCNN), a multi-layer perceptron (MLP), and a Transformer Encoder. The GCNN models the topological relationships between LW nodes and sensors. The MLP reduces data dimensionality while preserving key information for subsequent analysis. Meanwhile, the Transformer Encoder, employing multi-head attention mechanisms, captures global time-series patterns in the LW signals. Additionally, the CFRP-former framework includes a parallel module that separately predicts both the size and location of damage. Experiments using only four sensors show that CFRP-former achieves high accuracy in detecting and quantifying damage, with robustness confirmed through ablation tests.

[formula omitted]Beach: Self-attention-based spatiotemporal network for skillful prediction of shoreline changes multiple days ahead

We developed a self-attention-based spatiotemporal network called alphaBeach that uses spatiotemporal representation learning for skillful prediction of shoreline changes multiple days ahead. The proposed model predicts the spatiotemporal position of the shoreline up to seven consecutive days in the future based on hydrodynamic forcing of ocean waves and tide data for the past 30 consecutive days. It is further divided into alphaBeach-w/oIC and alphaBeach-w/IC depending on whether or not the beach state of the antecedent historical shoreline information is used as the initial condition. alphaBeach-w/oIC, which does not incorporate this information, learns the sequential relationship between hydrodynamic forcing and shoreline to estimate overall trends of shoreline changes including seasonal oscillation from the point in time after model training, given only the ocean waves and tides. alphaBeach-w/IC does incorporate antecedent historical shoreline information to greatly enhance its predictive accuracy of shoreline progradation, retreat, and beach rotation for short-term time scales and for extreme storm events. The proposed model was applied to Tairua Beach, New Zealand, and it demonstrated superior predictive accuracy compared to previous methods and matched current understanding of accretion-dominated and oscillation-dominated shoreline changes.

Self-Supervised Spatiotemporal Clustering of Vehicle Emissions With Graph Convolutional Network.

Spatiotemporal clustering of vehicle emissions, which reveals the evolution pattern of air pollution from road traffic, is a challenging representation learning task due to the lack of supervision. Some recent work building upon graph convolutional network (GCN) models the intrinsic spatiotemporal correlations among the nodes in road networks as graph representations for clustering. However, these existing methods ignore the interactions between spatial and temporal variations in vehicle emissions, resulting in incomplete descriptions and inaccurate detection of the evolution pattern of air pollution. To address this issue, this article proposes a two-way self-supervised spatiotemporal representation learning scheme, in which the temporal and spatial features are progressively learned in a mutually reinforced manner. Our proposed method is based on the observation that though the variation in vehicle emissions in the road network is consistent in the spatial and temporal domains, its expression is more distinct in temporal sequences. To this end, the input emission data are first projected into an initial temporal representation space spanned by the captured features from a pretrained BiLSTM network. Then the generated distribution of temporal features is used to construct an objective constraint for high-purity clustering through a two-way self-supervised mechanism, which is leveraged as a constraint for the feature clustering of a GCN. Furthermore, to eliminate the initial errors, a joint optimization scheme is presented to generate the decoupled clustering results through the progressive refinement of representation and clustering. Our proposed method is evaluated on the traffic emission dataset of Xian city in 2020, and the experimental results have demonstrated the superiority against the state-of-the-art.

Energize sustainability: EnSAF for sustainability aware, software intensive energy management systems

Context:India’s coal use for electricity jumped 13% in 2021–22. Energy management systems (EnMS) are seen as a solution, but only sustainable EnMS can have a discernable impact on the carbon footprint and the Return On Investment (ROI). Objective:Designing a software-intensive sustainable energy management system requires considering technical, environmental, social, and economic factors. This helps evaluate an EnMS’s overall impact and improve its design. We proposed EnSAF for efficient utilization of the energy incurred for the design of sustainability-aware EnMSs. Method:In this work, EnMSs in diverse use cases were selected and analyzed in terms of technical, social, environmental, and economic dimensions of sustainability in collaboration with various stakeholders. The set of application-specific design concerns and Quality Attributes (QAs) were addressed by the Sustainability Assessment Framework (SAF) toolkit. The resultant SAF instances of each EnMS, derived through the analysis and discussion with the stakeholders, were then analyzed to advocate the DMs and SQ model for generic EnMSs. Results:This study demonstrated the following outcomes (i) technical concerns dominate the existing EnMSs (ii) integration of renewable energy resources reduces dependency to the main power grid and nurtures a sustainable environment by diminishing carbon footprint, and minimizing payback time, in the economic dimension; (iii) extant definitions of quality attributes need significant scrutiny and updates apropos of objectives of EnMSs Conclusion:The SAF toolkit was found to be deficient in the representation of relevant design concerns and quality attributes concomitant with sustainable EnMS. Prevailing DMs are inept to factor in stakeholder’s concerns, as the model is ill-equipped to account for spatio-temporal representation of QAs. Pursuant to the insights from the 4 SAF instances, a generic framework, EnSAF, is proposed to tackle the relevant concerns apropos of EnMS sustainability. This work proposed a representation of DMs in the SAF toolkit specifically for sustainability-aware EnMS.

Signature of climate dynamics on hydrological drought dynamics: A qualitative analysis

Understanding the complexity of climate dynamics is of paramount importance, because climate variability and change significantly affect water cycles and ecosystems. However, recurrent hydrological droughts that have been observed every 10 years on the two primary tributaries of the Congo River (the Ubangui and Kasaï Rivers) since the advent of the 1969 drought lack a plausible explanation for variability or climate change. This study proposes a plausible explanation for recurring hydrological droughts. Given the low rate of human activity and vegetation cover evolution in the Congo Watershed, we propose that climate dynamics play a crucial role in hydrological drought dynamics. By applying the Bayesian approach to a gridded precipitation database, we obtained posterior probability maps for each annual time step during our observational period (1940–2020). This provided a spatiotemporal representation of the areas affected by climatic disturbances, unlike previous studies that were limited to a spatial representation of the temporal location of the disturbances. Our qualitative analysis of the maximum intensity of the climate disturbance signal (CPS) revealed an average cycle of 10 years and eight months of signal migration. However, we observed that every 10 years since the advent of the drought during 1969, the hydrological drought occurrence dates coincided with CPS migration dates. This highlights the influence of this cycle on the hydrological drought dynamics. Nevertheless, because of the monthly scale involved in the propagation time from meteorological to hydrological drought, the error in this cycle was considered to be the propagation time of disturbances. Therefore, we recommend that future research should focus on estimating this time to test this hypothesis. This study underscores the significance of the cyclic dynamics of lengthy transient processes in understanding hydrological drought dynamics.

Spatiotemporal implicit neural representation as a generalized traffic data learner

Spatiotemporal Traffic Data (STTD) measures the complex dynamical behaviors of the multiscale transportation system. Existing methods aim to reconstruct STTD using low-dimensional models. However, they are limited to data-specific dimensions or source-dependent patterns, restricting them from unifying representations. Here, we present a novel paradigm to address the STTD learning problem by parameterizing STTD as an implicit neural representation. To discern the underlying dynamics in low-dimensional regimes, coordinate-based neural networks that can encode high-frequency structures are employed to directly map coordinates to traffic variables. To unravel the entangled spatial–temporal interactions, the variability is decomposed into separate processes. We further enable modeling in irregular spaces such as sensor graphs using spectral embedding. Through continuous representations, our approach enables the modeling of a variety of STTD with a unified input, thereby serving as a generalized learner of the underlying traffic dynamics. It is also shown that it can learn implicit low-rank priors and smoothness regularization from the data, making it versatile for learning different dominating data patterns. We validate its effectiveness through extensive experiments in real-world scenarios, showcasing applications from corridor to network scales. Empirical results not only indicate that our model has significant superiority over conventional low-rank models, but also highlight that the versatility of the approach extends to different data domains, output resolutions, and network topologies. Comprehensive model analyses provide further insight into the inductive bias of STTD. We anticipate that this pioneering modeling perspective could lay the foundation for universal representation of STTD in various real-world tasks. PyTorch implementations of this project is publicly available at:https://github.com/tongnie/traffic_dynamics.

Rethinking prediction-based video anomaly detection from local–global normality perspective

Video anomaly detection (VAD) has been intensively studied for years because of its potential applications in intelligent video systems. Prediction-based VAD methods have gained a lot of attention in recent years due to their simplicity and effectiveness, which learn normality through a frame prediction tasks. However, we find that they often consider only local or global normality in the temporal dimension thus are difficult to generalize for various scenarios. Specifically, some of them focus on learning local spatiotemporal representations from consecutive frames to enhance the representation for normal events while the other methods are devoted to memorizing prototypical normal patterns of whole training videos to weaken the generalization for anomalies. In this paper, we rethink these two types of methods and find that the former methods are difficult to adapt to simple scenarios due to powerful representation allows these methods to represent some anomalies and causes miss detection while the latter struggle with complex scenarios due to the limitations for diverse normal patterns and causes false alarm. To verify our discoveries, we design a two-branch model, Local–Global Normality Network (LGN-Net), to simultaneously learn local and global normality. We verified the effect of local and global normality on the VAD model by ablation study and find that the model can be more adaptive to different scenarios when accounting for both normality. The code is available online: https://github.com/Myzhao1999/LGN-Net.

Kant and his Critique of Leibniz

According to one line of interpretation, in the Critique of Pure Reason, Immanuel Kant argues that Gottfried Wilhelm Leibniz conflated two kinds of cognitive capacities and their respective types of mental representations: (1) sensibility and sensible or spatiotemporal representations and (2) the understanding and intellectual or conceptual representations. According to this interpretation, Leibniz held that sensibility is not distinct from the understanding, and sensible representations are confused and reducible to distinct intellectual representations. However, this interpretation fails against the more charitable thesis that Leibniz believed the two cognitive capacities and their representations are distinct, yet all sensible representations are necessarily confused. Instead, I argue that Kant’s criticism concerns a different distinction: general logic, the rules of thought in general, and transcendental logic, the rules governing thought of sensible representations. Additionally, I argue for a second part of Kant’s criticism: the Copernican Revolution. By providing an account of mathematics that depends on objects of possible knowledge conforming to our cognitive capacities, Kant eliminates the possibility of a Leibnizian metaphysical project that allegedly aims beyond such objects. A Leibnizian would have to provide an account of the success of mathematics solely in terms of general logic or deny it altogether.

Contrastive learning of shared spatiotemporal EEG representations across individuals for naturalistic neuroscience

Neural representations induced by naturalistic stimuli offer insights into how humans respond to stimuli in daily life. Understanding neural mechanisms underlying naturalistic stimuli processing hinges on the precise identification and extraction of the shared neural patterns that are consistently present across individuals. Targeting the Electroencephalogram (EEG) technique, known for its rich spatial and temporal information, this study presents a framework for Contrastive Learning of Shared SpatioTemporal EEG Representations across individuals (CL-SSTER). CL-SSTER utilizes contrastive learning to maximize the similarity of EEG representations across individuals for identical stimuli, contrasting with those for varied stimuli. The network employs spatial and temporal convolutions to simultaneously learn the spatial and temporal patterns inherent in EEG. The versatility of CL-SSTER was demonstrated on three EEG datasets, including a synthetic dataset, a natural speech comprehension EEG dataset, and an emotional video watching EEG dataset. CL-SSTER attained the highest inter-subject correlation (ISC) values compared to the state-of-the-art ISC methods. The latent representations generated by CL-SSTER exhibited reliable spatiotemporal EEG patterns, which can be explained by properties of the naturalistic stimuli. CL-SSTER serves as an interpretable and scalable framework for the identification of inter-subject shared neural representations in naturalistic neuroscience.

Adversarial Spatiotemporal Contrastive Learning for Electrocardiogram Signals.

Extracting invariant representations in unlabeled electrocardiogram (ECG) signals is a challenge for deep neural networks (DNNs). Contrastive learning is a promising method for unsupervised learning. However, it should improve its robustness to noise and learn the spatiotemporal and semantic representations of categories, just like cardiologists. This article proposes a patient-level adversarial spatiotemporal contrastive learning (ASTCL) framework, which includes ECG augmentations, an adversarial module, and a spatiotemporal contrastive module. Based on the ECG noise attributes, two distinct but effective ECG augmentations, ECG noise enhancement, and ECG noise denoising, are introduced. These methods are beneficial for ASTCL to enhance the robustness of the DNN to noise. This article proposes a self-supervised task to increase the antiperturbation ability. This task is represented as a game between the discriminator and encoder in the adversarial module, which pulls the extracted representations into the shared distribution between the positive pairs to discard the perturbation representations and learn the invariant representations. The spatiotemporal contrastive module combines spatiotemporal prediction and patient discrimination to learn the spatiotemporal and semantic representations of categories. To learn category representations effectively, this article only uses patient-level positive pairs and alternately uses the predictor and the stop-gradient to avoid model collapse. To verify the effectiveness of the proposed method, various groups of experiments are conducted on four ECG benchmark datasets and one clinical dataset compared with the state-of-the-art methods. Experimental results showed that the proposed method outperforms the state-of-the-art methods.

Objective representative flow field selection for tidal array layout design

The representation of flow across influential spatiotemporal scales introduces a challenge when micro-siting tidal stream turbine arrays. Robust representative approximations could accelerate design optimisation, yet there is no consensus on what defines the most appropriate flow conditions. We summarise existing approaches to representative flow field selection for array optimisation and propose an objective-driven process. The method curates a subset of flow fields that best captures relevant dynamics, enabling the streamlined representation of tidal cycles. To demonstrate the method, we consider flow modelling data in the Inner Sound of the Pentland Firth, Scotland, UK. We examine the impact of flow field inputs to array design through comparative analyses using a heuristic array optimisation process. Results indicate notable sensitivity of the turbine layout to the flow conditions selected. For the case study presented, our method led to 4%–5% energy yield prediction improvements relative to use of simple time-interval based approaches and up to 2% improvement against using peak flow fields; these can be pivotal margins to secure feasibility by developers. We also find that using the data associated with a single monitored point across the array for flow field selection can lead to sub-optimal results, emphasising the need for accurate spatiotemporal representation.

MeST-Former: Motion-enhanced Spatiotemporal Transformer for generalizable Deepfake detection

The rise of Deepfake technology has sparked significant concerns due to its potential for misuse and malicious manipulation of multimedia content. Various detection approaches aimed at detecting Deepfake videos have been proposed, mostly relying on the identification of spatial and temporal artifacts. However, due to the different contexts of source images and the variety of generation techniques, current Deepfake detection methods usually perform well on training datasets, and yet generalize poorly to those unseen identities in new datasets. This issue is widely known as the generalization challenge of Deepfake detection. To address this challenge, this paper proposes an advanced spatiotemporal Deepfake video detector, named Motion-enhanced Spatiotemporal Transformer (MeST-Former). MeST-Former is based on the spatiotemporal modeling capacity of the video Swin Transformer. The spatial and temporal features are obtained from the RGB and motion images, respectively. To enhance the generalization ability of MeST-Former to unseen identities in unseen datasets, the ID-related components in the spatial and temporal features are detached. Specifically, MeST-Former adopts the newly proposed Identity-Decoupling Attention (IDC-Att) module to disentangle the ID-related and ID-unrelated components. Only the ID-unrelated components are used to construct more generalizable spatiotemporal representations. This process makes the constructed spatiotemporal features identity-agnostic and more generalizable to unseen identities. We conducted extensive experiments to evaluate the performance of the MeST-Former. Our results indicate that MeST-Former achieves accurate and generalizable Deepfake detection performance. Notably, MeST-Former also demonstrates high efficacy in detecting AI-animated talking-head videos.