Temporal Stream Research Articles

Multimodal Information Fusion Dynamic Target Recognition for Autonomous Driving

With the continuous development of autonomous driving technology, realizing high-precision road obstacle detection is crucial to ensure traffic safety and driving experience. However, traditional obstacle detection methods often perform poorly in complex driving scenarios, such as obstacle movement and occlusion. To cope with this problem, this study proposes a road obstacle detection method based on a two-stream convolutional neural network model, aiming to overcome the limitations of traditional methods in capturing spatiotemporal features and handling complex situations. Our research approach is based on the following innovations: First, we introduce a dual-stream convolutional neural network structure, where one stream is used to extract spatial features from the contour information of the obstacle frames, and the other stream extracts temporal features from the temporal stream information. This dual-stream structure can fully capture the appearance and dynamic information of the obstacles, thus improving detection accuracy. Second, we design a feature fusion module to fuse the two features to obtain richer obstacle features. In addition, we propose a new loss function, i.e. clustering loss, for better optimizing the model training process, reducing intra-class variation, and increasing inter-class differences, thus improving the generalization performance of the model. In the experimental section, we conducted extensive experimental analysis using Citycapes and BDD100K datasets. The experimental results show that our model achieves significant performance improvement compared to both the traditional convolutional neural network method and the YOLOv5 method in a variety of scenarios such as obstacle stationary, obstacle moving, and obstacle with occlusion. Specifically, our method improves the recognition rate up to 4.8% to 14.5% on the Citycapes dataset and 6.5% to 12.8% on the BDD100K dataset, respectively, under different scenarios. In addition, our model also exhibits more advantages on small datasets, showing higher generalization ability and robustness.

A Dual-Stream Fusion Network for Human Energy Expenditure Estimation with Wearable Sensor

With the increasing awareness of health, using wearable sensors to monitor individual activities and accurately estimate energy expenditure has become a current research focus. However, existing research encounters challenges including low estimation accuracy, a deficiency of frequency domain features, and difficulty in integrating time domain and frequency domain features. To address these issues, we propose an innovative framework called the Dual-Stream Fusion Network (DSFN). This framework combines the Time Domain Encoding (TDE) module, the Frequency Domain Hierarchical-Split Encoding (FDHSE) module, and a Two-Stage Feature Fusion (TSF) module. Specifically, the temporal stream of the framework employs the TDE module to capture deep temporal features that reflect the complex dynamic variations in time-series data. The frequency domain stream introduces the FDHSE module, which extracts frequency domain features using a multi-level, multi-scale approach, ensuring a comprehensive and diverse representation of frequency information. Through this dual-stream architecture, our model effectively learns both time and frequency domain features, addressing the limitations of frequency domain features observed in prior studies. Additionally, we propose the TSF module to fully integrate time and frequency domain features, effectively overcoming the challenge of fusing these two types of features. We conducted experiments on two public datasets, namely the GOTOV dataset (elderly people) and the JSI dataset (young people). Experimental results demonstrate that our method achieves excellent performance across different age groups. Compared to the baseline models, the proposed DSFN significantly improves the accuracy of human energy expenditure estimation.

Medical tourism in South East Asia: science mapping of present and future trends

PurposeThis study evaluates the knowledge structure of medical tourism within the geographical context of South East Asia. This region is one of the growing economic powerhouses in the world, and tourism activities have contributed a lot to its advancement.Design/methodology/approachApplying a science mapping technique using bibliometric analysis, the current and emerging themes and future trends are analyzed using bibliographic coupling and co-word analysis.FindingsFindings show that current trends produced four themes: Fundamentals of medical tourism in Southeast Asia, determinants of tourist medical tourism visits, quality of medical and health service in Southeast Asia and impact of medical tourism on national economic growth. The future trends also produced four themes related to navigating excellence in medical tourism, medical tourism and economic growth, service quality in medical tourism services and accredited destinations in the globalized era of medical tourism.Research limitations/implicationsThis study is relevant to all stakeholders, operators and local communities in Southeast Asia tourism destinations to provide the best medical tourism with the best quality service and technologies.Originality/valueThis study fills the gap by performing a bibliometric approach to reviewing medical tourism in Southeast Asia using a science mapping technique. Crucial themes are produced through topological and temporal streams that provide critical insight for future developments in medical tourism in the region.

A Video Action Recognition Method via Dual-Stream Feature Fusion Neural Network with Attention

Video action recognition is a technique for automatically determining the category of a video action. It is necessary to design an efficient video action recognition algorithm to predict video labels. This work proposes a video action recognition model based on dual-stream information fusion with attention mechanisms (DSIFAM), which consists of three different sub-modules. First, this proposes an improved keyframe extraction method (IKFE). Based on K-means clustering, this uses convolutional features to calculate the similarity between video frames instead of pixel points. After obtaining preliminary clustering results, the method performs secondary optimization to obtain more representative keyframes. Second, this proposes a video action recognition model based on dual-stream information fusion (DSIF). The method introduces ConvLSTM in the spatial stream and uses P3D instead of the original convolutional network in the temporal stream, which can better extract spatial-temporal information and improve the classification performance. Third, this designs a multi-scale attention mechanism (MSAM) to enhance the feature extraction stage and obtain higher quality classification features. The resulting features are more prominent and have stronger representation capabilities. Finally, this work conducts systematic experiments on different datasets, the results verify the superiority of DSIFAM for video action recognition.

Mixed Resolution Network with hierarchical motion modeling for efficient action recognition

The dual-stream architecture is frequently employed for learning diverse features from videos. This paper introduces a novel Mixed Resolution Network (MixRes) for processing inputs with hybrid spatiotemporal resolutions, namely high-spatial and low-temporal resolution input, as well as low-spatial and high-temporal resolution input. The utilization of mixed spatiotemporal resolutions not only facilitates the independent emphasis of the two streams on appearance and motion encoding but also diminishes the computational burden. Furthermore, by leveraging the characteristics of neural networks with multiple layers, the temporal stream in the proposed network is divided into different steps to capture short-term and long-term motion information. Finally, we design a Temporal Multiscale Motion Excitation (TMME) module, which enhances the motion-related channels of the video representation by employing multiscale temporal differences. We conduct extensive experiments on multiple action recognition benchmarks, including Something-Something V1 & V2 and Kinetics-400. The outcomes validate that the proposed method achieves superior action recognition performance with low computational cost as compared to the state-of-the-art methods.

Edge-Enhanced TempoFuseNet: A Two-Stream Framework for Intelligent Multiclass Video Anomaly Recognition in 5G and IoT Environments

Surveillance video analytics encounters unprecedented challenges in 5G and IoT environments, including complex intra-class variations, short-term and long-term temporal dynamics, and variable video quality. This study introduces Edge-Enhanced TempoFuseNet, a cutting-edge framework that strategically reduces spatial resolution to allow the processing of low-resolution images. A dual upscaling methodology based on bicubic interpolation and an encoder–bank–decoder configuration is used for anomaly classification. The two-stream architecture combines the power of a pre-trained Convolutional Neural Network (CNN) for spatial feature extraction from RGB imagery in the spatial stream, while the temporal stream focuses on learning short-term temporal characteristics, reducing the computational burden of optical flow. To analyze long-term temporal patterns, the extracted features from both streams are combined and routed through a Gated Recurrent Unit (GRU) layer. The proposed framework (TempoFuseNet) outperforms the encoder–bank–decoder model in terms of performance metrics, achieving a multiclass macro average accuracy of 92.28%, an F1-score of 69.29%, and a false positive rate of 4.41%. This study presents a significant advancement in the field of video anomaly recognition and provides a comprehensive solution to the complex challenges posed by real-world surveillance scenarios in the context of 5G and IoT.

3D Convolutional Neural Network-Based Multi-Parameter Video Quality Assessment Model on Cloud Platforms

In light of the rapid advancements in big data and artificial intelligence technologies, the trend of uploading local files to cloud servers to mitigate local storage limitations is growing. However, the surge of duplicate files, especially images and videos, results in significant network bandwidth wastage and complicates server management. To tackle these issues, we have developed a multi-parameter video quality assessment model utilizing a 3D convolutional neural network within a video deduplication framework. Our method, inspired by the analytic hierarchy process, thoroughly evaluates the effects of packet loss rate, codec, frame rate, bit rate, and resolution on video quality. The model employs a two-stream 3D convolutional neural network to integrate spatial and temporal streams for capturing video distortion details, with a coding layer configured to remove redundant distortion information. We validated our approach using the LIVE and CSIQ datasets, comparing its performance against the V-BLIINDS and VIDEO schemes across different packet loss rates. Furthermore, we simulated the client-server interaction using a subset of the dataset and assessed the scheme's time efficiency. Our results indicate that the proposed scheme offers a highly efficient solution for video quality assessment.

Multi-motion sensor behavior based continuous authentication on smartphones using gated two-tower transformer fusion networks

With continuous authentication, smartphones can constantly calculate authentication scores to verify a user's identity while interacting with the smartphones after logging in. Among various behavioral biometrics based continuous authentication methods, motion dynamics biometrics based methods draw a lot of attention, which relies on potential differences in signals from smartphone built-in motion sensors to provide an implicit and secure authentication. However, existing motion dynamics based methods not only fail to model correlations between motion sensors with multiple channels but also ignore temporal patterns contained in each channel. In this paper, we develop a two-tower neural networks based continuous authentication system TNNAuth based on the built-in multi-motion sensors on smartphones. Specifically, TNNAuth divides the raw multi-motion sensor data into channel and temporal stream data, and proposes a gated two-tower transformer fusion network to capture both time-over-channel and channel-over-time motion patterns. To evaluate TNNAuth, we collect a large-scale multi-motion sensor dataset in the unconstrained real life. Extensive experiments demonstrate that TNNAuth can provide the lowest authentication error rate, with a false-acceptance rate of 8.3% and a false-rejection rate of 9.4%.

TSDTVOS: Target-guided spatiotemporal dual-stream transformers for video object segmentation

Video object segmentation automatically separates the interested objects from the background across a video sequence and was an active research area in recent years. The crucial challenge lies in investigating an effective architecture to fully exploit spatiotemporal correlation in a given video sequence for achieving accurate segmentation results. In this paper, we propose a novel semi-supervised Transformer-based framework called Target-guided Spatiotemporal Dual-stream Transformers (TSDT) with two separate streams to enable effective spatiotemporal context propagation. Technically, the temporal stream is used to aggregate rich temporal cues from past frames, while the spatial stream is trained to encode object location and appearance information stored in the current frame. To compress and integrate temporal features, a target guidance block (TGB) is designed to retrieve target information in the past video flow under the guidance of the current frame. The experimental results on video object segmentation benchmarks demonstrate the feasibility and effectiveness of the proposed framework. Codes and trained models are available at https://github.com/zhouweii234/TSDTVOS.

Abstract P1119: Deep Learning-based Cardiac Microphysiological Systems For Studying Reentry Arrhythmia

Arrhythmias can be caused by reentry and focal activities, leading to a disruption of the normal sinus rhythm. Understanding these mechanisms and their contributions to arrhythmias in individual patients is crucial for developing effective treatments. To achieve this, in vitro human models based on induced pluripotent stem cell-derived cardiomyocytes have been developed to replicate arrhythmogenic activities and identify pathophysiological mechanisms specific to patients. However, the current methods for analyzing electrophysiology (EP) videos are limited and subjective, making the interpretation of clinical EP videos challenging. To address these limitations, we have developed a deep cardiac EP system called cEP-Net. The system comprises a cardiac microphysiological system designed for reproducing various forms of reentry and focal activities in vitro and a deep learning-based EP video classifier with a spatial and temporal streams architecture. The deep EP classifier analyzes cardiac EP videos, identifies arrhythmogenic activities, and categorizes videos into four categories based on their characteristics: (1) localized activities, (2) normal linear propagation, (3) focal activity, and (4) reentrant activity. We pre-trained cEP-Net using 111,110 annotated in vitro EP videos produced by a cardiac microphysiological system and evaluated its performance for in vitro and in vivo clinical settings. Our pre-trained cEP-Net demonstrated high Kappa (91.0%) and concordance rate (94.7%) for all categories. cEP-Net visualized in vitro EP videos in UMAP domains and successfully clustered them into four categories. Additionally, cEP-Net successfully captured the development and evolution of reentry during early in vitro cardiac tissue formation and the initiation of focal activities in the HCN4 (a pacemaker protein)-expressing tissues. We also used cEP-Net to quantify the arrhythmogenic risks of cardiac agents and found that cEP-Net could capture Flecainide-induced vulnerability to reentry. Lastly, we applied cEP-Net to analyze electroanatomical mapping videos in vivo. The cEP-Net demonstrated its capability to identify phase singularities in hearts with atrial fibrillation (AF) and capture the EP patterns of persistent AF hearts.

The impacts of human-induced disturbances on spatial and temporal stream water quality variations in mountainous terrain: A case study of Borcka Dam Watershed

Unaltered watersheds with natural vegetation cover (forest, grasslands, etc.) provide several ecological benefits in addition to providing freshwater, controlling water levels, and supporting flourishing streamside ecosystems. However, as in many watersheds in the World, the research area in this study, the Borcka Dam Watershed (BDW), has been affected by many human-induced disturbances affecting a wide area of forest and grassland areas as well as soil and water resources. Therefore, the objective of this study was to assess and evaluate the possible effects of anthropogenic disturbances, particularly on annual changes in water discharge, some water quality parameters, and total suspended sediment (TSS) amounts in the main streams of four sub-watersheds (Fabrika, Godrahav, Hatila, and Murgul) and the reservoir of the dam. In addition, we intend to confirm that land use change and/or transformation play a significant role in influencing stream water quality. The YSI/Professional-Plus, a portable water quality measurement device, was used to determine the amounts of pH, dissolved oxygen (DO), total dissolved substance (TDS), ammonium (NH4–N), nitrate (NO3–N), salinity, electrical conductivity (EC), and temperature besides measuring discharge and total suspended sediments (TSS) from a total of 27 sampling points in the field. Although the results revealed that the annual mean values of all water quality parameters for all four streams were mostly in good condition, for some time and points of the measurements, several parameters were found to be above the official water quality standards due to the intensive aforementioned anthropogenic activities, particularly in the stream waters of Murgul (e.g. pH and TSS being 10,84 and 236 mg/L, respectively) and Fabrika (e.g. EC of 412 μs/cm; DO of 4.44 mg/L; 14 ml of NO3–N) sub-watersheds. These outcomes indicate that these two sub-watersheds have been impacted more severely by the human-induced disturbances compared to Hatila and Godrahav sub-watersheds.

A Bi-Stream hybrid model with MLPBlocks and self-attention mechanism for EEG-based emotion recognition

Due to the instability and complex distribution of electroencephalography (EEG) signals and the great cross-subject variations, exploiting valuable and discriminative emotional information from EEG is still a significant challenging issue. In this paper, we propose Bi-Stream Multilayer Perceptron – Self-Attention Mixer (BiSMSM), a novel model for EEG-based emotion recognition. The proposed model consists of two streams: the spatial stream and the temporal stream. BiSMSM jointly captures the useful information from temporal, spatial, local and global angles, aiming to encode more discriminative features describing emotions. The spatial stream focuses on the spatial information, while the temporal stream concentrates on the correlation information in the time domain. The structures of the two streams are similar, either of which contains a Multilayer Perceptron (MLP) based module to extract the regional in-channel and cross-channel information. The MLP-based module is followed by a self-attention mechanism module to explore the global signal correlations. Finally, the subject-independent experiments on the public benchmark datasets DEAP and DREAMER have demonstrated the advantage of our model over the related advanced approaches. Specifically, BiSMSM obtains an accuracy of 63.10% for valence classification and 61.89% for aoursal classification on DEAP, and 61.88% for valence classification and 64.25% for arousal classification on DREAMER.

Blood flow characterization in nailfold capillary using optical flow-assisted two-stream network and spatial-temporal image

The blood flow velocity in the nailfold capillary is an important indicator of the status of microcirculation. The conventional manual processing method is both laborious and prone to human artifacts. A feasible way to solve this problem is to use machine learning to assist in image processing and diagnosis. Inspired by the Two-Stream Convolutional Networks, this study proposes an optical flow-assisted two-stream network to segment nailfold blood vessels. Firstly, we use U-Net as the spatial flow network and the dense optical flow as the temporal stream. The results show that the optical flow information can effectively improve the integrity of the segmentation of blood vessels. The overall accuracy is 94.01 %, the Dice score is 0.8099, the IoU score is 0.6806, and the VOE score is 0.3194. Secondly, The flow velocity of the segmented blood vessel is determined by constructing the spatial-temporal (ST) image. The blood flow velocity evaluated is consistent with the typical blood flow speed reported. This study proposes a novel two-stream network for blood vessel segmentation of nailfold capillary images. Combined with ST image and line detection method, it provides an effective workflow for measuring the blood flow velocity of nailfold capillaries.

Temporal integration is a robust feature of perceptual decisions.

Making informed decisions in noisy environments requires integrating sensory information over time. However, recent work has suggested that it may be difficult to determine whether an animal's decision-making strategy relies on evidence integration or not. In particular, strategies based on extrema-detection or random snapshots of the evidence stream may be difficult or even impossible to distinguish from classic evidence integration. Moreover, such non-integration strategies might be surprisingly common in experiments that aimed to study decisions based on integration. To determine whether temporal integration is central to perceptual decision making, we developed a new model-based approach for comparing temporal integration against alternative 'non-integration' strategies for tasks in which the sensory signal is composed of discrete stimulus samples. We applied these methods to behavioral data from monkeys, rats, and humans performing a variety of sensory decision-making tasks. In all species and tasks, we found converging evidence in favor of temporal integration. First, in all observers across studies, the integration model better accounted for standard behavioral statistics such as psychometric curves and psychophysical kernels. Second, we found that sensory samples with large evidence do not contribute disproportionately to subject choices, as predicted by an extrema-detection strategy. Finally, we provide a direct confirmation of temporal integration by showing that the sum of both early and late evidence contributed to observer decisions. Overall, our results provide experimental evidence suggesting that temporal integration is an ubiquitous feature in mammalian perceptual decision-making. Our study also highlights the benefits of using experimental paradigms where the temporal stream of sensory evidence is controlled explicitly by the experimenter, and known precisely by the analyst, to characterize the temporal properties of the decision process.

MHAMD-MST-CNN: multiscale head attention guided multiscale density maps fusion for video crowd counting via multi-attention spatial-temporal CNN

ABSTRACT Video-based crowd counting and density estimation (CCDE) is vital for crowd monitoring. The existing solutions lack in addressing issues like cluttered background and scale variation in crowd videos. To this end, a multiscale head attention-guided multiscale density maps fusion for video-based CCDE via multi-attention Spatial-Temporal CNN (MHAMD-MST-CNN) is proposed. The MHAMD-MST-CNN has three modules: a multi attention spatial stream (MASS), a multi attention temporal stream (MATS), and a final density map generation (FDMG) module. The spatial head attention modules (SHAMs) and temporal head attention modules (THAMs) are designed to eliminate the background influence from the MASS and the MATS, respectively, by mapping the multiscale spatial or temporal features to head maps. The multiscale de-backgrounded features are utilised by the density map generation (DMG) modules to generate multiscale density maps to deal with scale variation due to perspective distortion. The multiscale density maps are fused and fed into the FDMG module to obtain the final crowd density map. The MHAMD-MST-CNN has been trained and validated on three publicly available benchmark datasets: the Venice, the Mall, and the UCSD. The MHAMD-MST-CNN provides competitive results as compared with the state-of-the-arts in terms of mean absolute error (MAE) and root mean squared error (RMSE).

3D network with channel excitation and knowledge distillation for action recognition.

Modern action recognition techniques frequently employ two networks: the spatial stream, which accepts input from RGB frames, and the temporal stream, which accepts input from optical flow. Recent researches use 3D convolutional neural networks that employ spatiotemporal filters on both streams. Although mixing flow with RGB enhances performance, correct optical flow computation is expensive and adds delay to action recognition. In this study, we present a method for training a 3D CNN using RGB frames that replicates the motion stream and, as a result, does not require flow calculation during testing. To begin, in contrast to the SE block, we suggest a channel excitation module (CE module). Experiments have shown that the CE module can improve the feature extraction capabilities of a 3D network and that the effect is superior to the SE block. Second, for action recognition training, we adopt a linear mix of loss based on knowledge distillation and standard cross-entropy loss to effectively leverage appearance and motion information. The Intensified Motion RGB Stream is the stream trained with this combined loss (IMRS). IMRS surpasses RGB or Flow as a single stream; for example, HMDB51 achieves 73.5% accuracy, while RGB and Flow streams score 65.6% and 69.1% accuracy, respectively. Extensive experiments confirm the effectiveness of our proposed method. The comparison with other models proves that our model has good competitiveness in behavior recognition.

Evaluation of Key Spatiotemporal Learners for Print Track Anomaly Classification Using Melt Pool Image Streams

Recent applications of machine learning in metal additive manufacturing (MAM) have shown great potential to resolve critical barriers to MAM's widespread adoption. Recent research on the topic highlights the significance of using melt pool signatures to predict defects on-the-fly. While high-fidelity melt pool image data has the potential to enable accurate predictions, hardly any work exists on the use of state-of-the-art spatiotemporal models to leverage the information embedded in the transient and sequential nature of the additive process. This work introduces and implements some of the major deep spatiotemporal learners that can be adapted to classify melt pool image streams from different materials, systems, and applications. In this regard, two-stream networks with a spatial and temporal stream, a recurrent spatial network, and a factorized 3D convolutional neural network were investigated herein. The generalization abilities of these models to perturbations in the melt pool image data are tested using data perturbation techniques that are grounded in palpable process situations. The implemented architectures exhibit the ability to learn the spatiotemporal features of the melt pool image sequences. However, only the Kinetics400 pre-trained SlowFast network, which belongs to the two-stream category, showed strong generalisation abilities to data perturbations.

Learning Spatiotemporal Graph Representations for Visual Perception Using EEG Signals.

Perceiving and recognizing objects enable interaction with the external environment. Recently, decoding brain signals based on brain-computer interface (BCI) that recognize the user's intentions by just looking at objects has attracted attention as a next-generation intuitive interface. However, classifying signals from different objects is very challenging, and in practice, decoding performance for visual perception is not yet high enough to be used in real environments. In this study, we aimed to classify single-trial electroencephalography signals evoked by visual stimuli into their corresponding semantic category. We proposed a two-stream convolutional neural network to increase classification performance. The model consists of a spatial stream and a temporal stream that use graph convolutional neural network and channel-wise convolutional neural network respectively. Two public datasets were used to evaluate the proposed model; (i) SU DB (a set of 72 photographs of objects belonging to 6 semantic categories) and MPI DB (8 exemplars belonging to two categories). Our results outperform state-of-the-art methods, with accuracies of 54.28 ± 7.89% for SU DB (6-class) and 84.40 ± 8.03% for MPI DB (2-class). These results could facilitate the application of intuitive BCI systems based on visual perception.

Dual-Stream Contrastive Learning for Channel State Information Based Human Activity Recognition.

WiFi-based human activity recognition (HAR) has been extensively studied due to its far-reaching applications in health domains, including elderly monitoring, exercise supervision and rehabilitation monitoring, etc. Although existing supervised deep learning techniques have achieved remarkable performances for these tasks, they are however data-hungry and hence are notoriously difficult due to the privacy and incomprehensibility of WiFi-based HAR data. Existing contrastive learning models, mainly designed for computer vision, cannot guarantee their performance on channel state information (CSI) data. To this end, we propose a new dual-stream contrastive learning model that can process and learn the raw WiFi CSI data in a self-supervised manner. More specifically, our proposed method, coined as DualConFi, takes raw WiFI CSI data as input and incorporates channel and temporal streams to learn highly-discriminative spatiotemporal features under a mutual information constraint using unlabeled data. We exhibit the effectiveness of our model on three publicly available CSI data sets in various experiment settings, including linear evaluation, semi-supervised, and transfer learning. We show that DualConFi is able to perform favourably against challenging baselines in each setting. Moreover, by studying the effects of different transform functions on CSI data, we finally verify the effectiveness of highly-discriminative features.

On Learning Mechanical Laws of Motion From Video Using Neural Networks

In computer vision, physics plays an important role in several applications. In this work, we teach a machine to detect the mechanical laws of motion of physical objects using video, and show how the results can be useful for computer vision tasks. We assume no prior knowledge of physics, beyond a temporal stream of bounding boxes. The problem is very difficult because a machine must learn not only a governing equation (e.g. projectile motion) but also the existence of governing parameters (e.g. velocities). We evaluate our ability to represent the physical laws of motion in video, such as the movement of a projectile or circular motion, in both real and constructed videos. These elementary tasks have textbook governing equations and enable ground truth verification of our approach. To establish the importance of the proposed method, we show a real-world use case in the domain of object tracking in confounding scenes, where existing state-of-the-art algorithms fail. Incorporating physics into computer vision not only serves the purpose of curiosity-driven research, but also provides an inductive bias for computer vision applications like object tracking.