Domain Generalization (DG) requires models to generalize across unseen data distributions. Kernel-based theory reveals a No-Free-Lunch problem: any model with a fixed representation is fundamentally sub-optimal for all possible shifts. While large ensembles mitigate this, they are computationally expensive and remain static once trained, inheriting the same theoretical limitation. We introduce MoE² (Mixture-of-Mixtures of Experts), a framework that uses a single frozen backbone to dynamically synthesize a bespoke adapter for each input, allowing it to continuously adapt its effective kernel. We provide a theoretical grounding for this process, proving our routing mechanism is a principled non-parametric estimator for the optimal Bayes mixture of experts. We derive a generalization bound that cleanly separates the router's estimation error from the reduction in a kernel-mismatch penalty achieved via synthesis. MoE² matches or exceeds state-of-the-art ensemble baselines on major DG benchmarks while using only a single, compact model. MoE² thus provides a theoretically-grounded and lightweight alternative to large-scale ensembles for robust domain generalization.

One-stream transformer-based trackers have shown remarkable success in single object tracking by jointly performing feature extraction and relation modeling. However, the update of temporal context, often propagated via temporal tokens, typically relies on general self-attention mechanisms within the transformer backbone. This paper introduces STTATrack, a novel framework that enhances one-stream tracking by explicitly leveraging the immediate spatial certainty from the current frame’s prediction score map to refine these propagated temporal query tokens. Our core contribution, the Score Temporal Token Attention (STTA) module, generates an embedding from the score map and employs a dual attention mechanism to facilitate bidirectional information flow between this spatial certainty embedding and the existing temporal tokens. This targeted refinement allows temporal tokens to be dynamically adapted based on the most current and spatially precise evidence, leading to improved adaptability and temporal consistency. STTATrack builds upon the ODTrack architecture and demonstrates significant performance improvements on the challenging GOT10k, OTB and UAV123 benchmarks, underscoring the efficacy of explicitly integrating current-frame spatial certainty into the temporal refinement loop.

We propose VOST, a novel segmentation-bytracking framework for semi-supervised Video Object Segmentation (VOS) that decouples the tracking and segmentation tasks to improve both accuracy and robustness. Our approach employs a state-of-the-art zero-shot tracker (SAMURAI) to generate bounding boxes of the target object in each frame. These cropped regions are then fed into a Vision Transformer (ViT)-based segmentation model trained to segment the object within the bounding box. By isolating the object from distracting background content and similar instances, our model eliminates ambiguity, simplifies the segmentation task, and requires no temporal memory to maintain object consistency. VOST is evaluated using three benchmark datasets, DAVIS16, DAVIS17, and SegTrackV2, achieving state-of-the-art performance. Specifically, VOST reaches an $\mathcal{M}$ score of 92.6 on DAVIS16, 88.9 on DAVIS17, and an $\mathcal{F}$-score of 0.929 on SegTrackV2, outperforming all previous methods. Additionally, VOST achieves a real-time inference speed of 20.6 FPS, offering an efficient and scalable solution for practical VOS applications. These results demonstrate the effectiveness of the segmentation-by-tracking paradigm and its potential as a competitive alternative to memory-based approaches.

In this paper, we design a deep learning-based convolutional autoencoder for channel coding and modulation. The objective is to develop an adaptive scheme capable of operating at various signal-to-noise ratios (SNR)s without the need for re-training. Additionally, the proposed framework allows validation by testing all possible codes in the codebook, as opposed to previous AI-based encoder/decoder frameworks which relied on testing only a small subset of the available codes. This limitation in earlier methods often led to unreliable conclusions when generalized to larger codebooks. In contrast to previous methods, our multi-level encoding and decoding approach splits the message into blocks, where each encoder block processes a distinct group of B bits. By doing so, the proposed scheme can exhaustively test 2Bpossible codewords for each encoder/decoder level, constituting a layer of the overall scheme. The proposed model was compared to classical polar codes and TurboAE-MOD schemes, showing improved reliability with achieving comparable, or even superior results in some settings. Notably, the architecture can adapt to different SNRs by selectively removing one of the encoder/decoder layers without re-training, thus demonstrating flexibility and efficiency in practical wireless communication scenarios.

Video object segmentation is a challenging task in computer vision. In this task, a learning model should be able to segment and track a specific set of objects through a frame sequence. This set of objects is given from the ground truth of the first frame in a sequence. To achieve domain generalization in this task, the learning model must be trained using a massive, labeled dataset that has almost all kinds of objects that can be seen in any frame sequence. However, such a dataset does not exist because the labeling process for this task is so expensive, as it requires per-pixel labeling for each frame in a given frame sequence. In this paper, we propose a novel learning technique and transformer architecture. This novel learning technique allows the model to learn effectively from a small, labeled dataset. Additionally, the novel architecture allows the model to produce segmentation output as a function of an input example, instead of relying on memorizing the representation of all objects to be segmented. The experiments show the superiority of the proposed model in comparison with state-of-the-art models by $10.6 \%$ when evaluated using out-of-domain frame sequences.

Large Language Models (LLMs) have demonstrated remarkable success in various natural language processing and software engineering tasks, such as code generation. The LLMs are mainly utilized in the prompt-based zero/few-shot paradigm to guide the model in accomplishing the task. GPT-based models are one of the popular ones studied for tasks such as code comment generation or test generation. These tasks are 'generative' tasks. However, there is limited research on the usage of LLMs for 'non-generative' tasks such as classification using the prompt-based paradigm. In this preliminary exploratory study, we investigated the applicability of LLMs for Code Clone Detection (CCD), a non-generative task. By building a mono-lingual and cross-lingual CCD dataset derived from CodeNet, we first investigated two different prompts using ChatGPT to detect Type-4 code clones in Java-Java and Java-Ruby pairs in a zero-shot setting. We then conducted an analysis to understand the strengths and weaknesses of ChatGPT in CCD. ChatGPT surpasses the baselines in cross-language CCD attaining an F1-score of 0.877 and achieves comparable performance to fully fine-tuned models for mono-lingual CCD, with an F1-score of 0.878. Also, the prompt and the difficulty level of the problems has an impact on the performance of ChatGPT. Finally, we provide insights and future directions based on our initial analysis1.

Code Clone Detection (CCD) is a software engineering task that is used for plagiarism detection, code search, and code comprehension. Recently, deep learning-based models have achieved an F1-Score (a metric used to assess classifiers) of $$\sim $$ 95% on the CodeXGLUE benchmark. These models require many training data, mainly fine-tuned on Java or C++ datasets. However, no previous study evaluates the generalizability of these models where a limited amount of annotated data is available. The main objective of this research is to assess the ability of the CCD models as well as few-shot learning algorithms for unseen programming problems and new languages (i.e., the model is not trained on these problems/languages). We assess the generalizability of the state-of-the-art models for CCD in few-shot settings (i.e., only a few samples are available for fine-tuning) by setting three scenarios: i) unseen problems, ii) unseen languages, iii) combination of new languages and new problems. We choose CodeNet and conduct our experiments on Java, C++, and Ruby languages. Then, we employ Model Agnostic Meta-learning (MAML), where the model learns a meta-learner capable of extracting transferable knowledge from the train set; so that the model can be fine-tuned using a few samples. Finally, we combine contrastive learning with MAML to further study whether it can improve the results of MAML. Our results show that the performance of the models drops $$\sim 50\%$$ for Java and $$\sim 20\%$$ for C++ and Ruby for unseen problems, which are then boosted by $$13\%$$ to $$24\%$$ F1 scores for Java and C++/Ruby, respectively when MAML is used. Similar observations are found for unseen languages and the third scenario. Though in case of third scenario (i.e., unseen problems and unseen languages) the scores are lower. Integrating contrastive learning with MAML did not help in boosting the performance more than what we could achieve with MAML. Our results open new avenues of research and the need to develop robust models for clone detection, in the settings we investigated here.

In few-shot classification, performing well on a testing dataset is a challenging task due to the restricted amount of labelled data available and the unknown distribution. Many previously proposed techniques rely on prototypical representations of the support set in order to classify a query set. Although this approach works well with a large, in-domain support set, accuracy suffers when transitioning to an out-of-domain setting, especially when using small support sets. To address out-of-domain performance degradation with small support sets, we propose Masked Embedding Modeling for Few-Shot Learning (MEM-FS), a novel, self-supervised, generative technique that reinforces few-shot-classification accuracy for a prototypical backbone model. MEM-FS leverages the data completion capabilities of a masked autoencoder to expand a given embedded support set. To further increase out-of-domain performance, we also introduce Rapid Domain Adjustment (RDA), a novel, self-supervised process for quickly conditioning MEM-FS to a new domain. We show that masked support embeddings generated by MEM-FS+RDA can significantly improve backbone performance on both out-of-domain and in-domain datasets. Our experiments demonstrate that applying the proposed technique to an inductive classifier achieves state-of-the-art performance on mini-imagenet, the CVPR L2ID Classification Challenge, and a newly proposed dataset, IKEA-FS. We provide code for this work at https://github.com/Brikwerk/MEM-FS.

Moving object detection is a core task in computer vision. However, existing deep learning-based moving object detection methods require a large number of labeled frames to achieve good generalization and performance. This paper proposes a novel deep learning network called FeSh-Net. This network can learn to extract an exemplar-based attention map using a few labeled frames, which guides the network to know which object is foreground and which is a background in the current frame. FeSh-Net is trained using a novel meta-learning technique to be able to segment moving objects from new unseen videos. The proposed network is evaluated using the benchmark CDNet. The results of the proposed FeSh-Net are compared with current state-of-the-art methods, and the results show that FeSh-Net outperforms the best reported state-of-the-art method by 4.4% on average. Additionally, FeSh-Net performs better than other methods when tested using new unseen videos.

Balancing the trade-off between real-time performance and accuracy in object tracking is a major challenge. In this paper, a novel dynamic policy gradient Agent-Environment architecture with Siamese network (DP-Siam) is proposed to train the tracker to increase the accuracy and the expected average overlap while performing in real-time. DP-Siam is trained offline with reinforcement learning to produce a continuous action that predicts the optimal object location. DP-Siam has a novel architecture that consists of three networks: an Agent network to predict the optimal state (bounding box) of the object being tracked, an Environment network to get the Q-value during the offline training phase to minimize the error of the loss function, and a Siamese network to produce a heat-map. During online tracking, the Environment network acts as a verifier to the Agent network action. Extensive experiments are performed on six widely used benchmarks: OTB2013, OTB50, OTB100, VOT2015, VOT2016 and VOT2018. The results show that DP-Siam significantly outperforms the current state-of-the-art trackers.

Portable ultrasound is increasingly used to assess jugular venous pressure (JVP) to approximate volume status in patients with congestive heart failure (CHF). Traditionally, increases in jugular venous pressure height signify increasing circulating blood volume. Emerging evidence, suggests that JVP correlates well with sonographic images of the internal jugular vein (IJV). This paper represents a preliminary investigation on the ability of cross-sectional area (CSA) of the IJV to measure relative changes in circulating blood volume. Fourteen healthy subjects had serial transverse ultrasound videos of their IJV captured while lying at five angles designed to simulate relative changes in blood volume. Ultrasound videos of the IJV were both manually and semi-automatically segmented, the CSA was measured, outliers were detected and removed, and Rotation Forest classifier was used to classify the data. By limiting the number of classes from five to three and removing outliers the accuracies improved from 59.50% to 91.05% and 62.74% to 91.89% for manual and semi-automatic segmentation, respectively. This pilot demonstrated that serial measurement of the CSA of the IJV in combination with machine learning techniques represents a viable opportunity to monitor changes in circulating blood volume in healthy subjects, setting the stage for a trial monitoring of patients with CHF.

Detecting moving objects has been well studied in the past due to its importance in computer vision applications. Nevertheless, in aerial imagery, the small sizes of moving objects and the camera motion present challenges to existing well-known detection methods. Most moving object detection methods have reported either high true detection rates associated with high false-detection rates, or low false-detection rates at the expense of lowering true detection rates. This paper proposes a novel method, Kinematic Regularization and Matrix Rank Optimization (KRMARO), to achieve high true-detection rates and reduce false-detection rates significantly. KRMARO introduces a formulation of the moving objects detection problem that integrates a novel kinematic regularization into the principal component pursuit. This formulation models moving objects as sparse, which is located in regions exhibiting unique kinematic properties, while the background is modeled as a low-rank matrix that is corrupted by this sparse. To solve the former formulation accurately, KRMARO proposes a solution based on the inexact Newton method and the inexact augmented Lagrange multiplier with backtracking behavior. The robustness of KRMARO is verified through testing on DARPA VIVID, UCF aerial action, and VIRAT aerial data sets and then comparing the results with relevant state-of-the-art methods.

Detection of relative changes in circulating blood volume is important to guide resuscitation and manage a variety of medical conditions, including sepsis, trauma, dialysis, and congestive heart failure. Recent studies have shown that estimates of circulating blood volume can be obtained from the cross-sectional area of the internal jugular vein (IJV) from ultrasound images. However, accurate segmentation and tracking of the IJV in ultrasound imaging is a challenging task and is significantly influenced by a number of parameters, such as the image quality, shape, and temporal variation. In this paper, we propose a novel adaptive polar active contour (Ad-PAC) algorithm for the segmentation and tracking of the IJV in ultrasound videos. In the proposed algorithm, the parameters of the Ad-PAC algorithm are adapted based on the results of segmentation in previous frames. The Ad-PAC algorithm is applied to 65 ultrasound videos captured from 13 healthy subjects, with each video containing 450 frames. The results show that spatial and temporal adaptation of the energy function significantly improves segmentation performance when compared with the current state-of-the-art active contour algorithms.

The detection of ground-moving objects in aerial videos has evolved over the years to handle more challenges such as large camera motion, the small size of the objects, and occlusion. Recently, aerial detection has been attempted using principal component pursuit (PCP) due to its superiority in detecting small moving objects. However, PCP-based detection methods generally suffer from high-false detections as well as high-computational loads. This paper presents a novel PCP-based detection method called kinematic regularization with local null space pursuit (KRLNSP) that drastically reduces false detections and the computational loads. KRLNSP models the background in an aerial video as a subspace that spans a low-dimension subspace while it models the moving objects as moving sparse. Accordingly, the detection is achieved by using multiple local null spaces and enhanced kinematic regularization. The multiple local null spaces allow real-time execution to nullify the background while preserving the moving objects unchanged. The kinematic regularization penalizes these moving objects to filter out false detections. The extensive evaluation of KRLNSP and relevant current state-of-the-art methods prove that the KRLNSP outperforms these methods (the true positive rate of KRLNSP is 98% and its false positive rate is 0.4%) and significantly reduces the computational loads (KRLNSP execution time is 0.3 s/frame).

Using Unmanned Aerial Vehicles (UAVs) or drones in Vehicular Ad-hoc Networks (VANETs) has started to attract attention. This paper proposes a mathematical framework to determine the minimum drone density (maximum separation distance between two adjacent drones) that stochastically limits the worst case for the vehicle-to-drone packet delivery delay. In addition, it proposes a drones-active service (DAS) that is added to the location service in a VANET to obtain the required number of active drones based on the current vehicular density while satisfying a probabilistic requirement for vehicle-to-drone packet delivery delay. Our goal is boosting VANET communications using infrastructure drones to achieve the minimum vehicle-to-drone packet delivery delay. We are interested in two-way highway VANET networks with low vehicular density. The simulation results show the accuracy of our mathematical framework and reflect the relation between the vehicle-to-drone packet delivery delay and the drone density.

Digital video stabilization is crucial in many applications such as object detection and tracking. It has been studied for decades yielding an extensive amount of literature in the field, however, current approaches suffer from either being computationally expensive or under-performing in terms of visual quality . In this paper, we present StableFlow, a novel real-time method that was inspired by the mass-spring-damper physical model. In StableFlow, a video frame is modelled as a mass suspended in each direction by a critically dampened spring and damper which can be fine-tuned to adapt with different shaking patterns. The proposed method is tested on video sequences that have different types of shakiness and diverse video contents. The obtained results are then compared to current state-of-the-art stabilization algorithms including Youtube stabilization and it is found that the proposed method significantly outperforms other algorithms in terms of visual quality while performing in real time.

Aerial imagery applications have gained a great interest especially in the area of comprehensive ground activities analysis. One of the key tasks in such applications is moving objects segmentation. Although many efforts have been presented in the literature that claim high true object detection rates, they still suffer from high false positive rates. This paper focuses on maintaining a high true positive detection rates while significantly reducing the false positive detection rates. To achieve this goal, this paper proposes a novel method that integrates matrix recovery concept with physical spring model to drastically reduce false detections. The proposed method segment all candidate moving objects by recovering the low rank matrix, which normally results high false positive detection. To reject false detections, each candidate moving object is modelled as a mass suspended by system of springs, such that the forces of springs attached to false detections is negligible whereas the forces of springs attached to a true moving object will be significant in response to the object motion. The results show that the proposed method, compared to other current state-of-the-art methods, achieved better true positive rates while drastically lowering the false positive rates.

Moving object detection is the first key step for many automated vision analysis applications. One of the major challenges to achieve accurate moving object detection is detecting moving objects in videos captured by moving camera platforms, also called active cameras, where both interest objects and background elements are moving. This paper presents a novel algorithm for moving objects detection from active cameras. The proposed method decomposes a video from an active camera into three components: background, moving objects, and transformation matrix between consecutive frames. The proposed method formulates the problem as a robust principle component analysis (PCA) problem (low rank matrix optimization problem) and solves it using inexact augmented Lagrange multiplier (IALM). In the proposed method, the background represents the low rank matrix, and the moving objects and transformation matrix are treated as added corruption. The robustness of the proposed method is demonstrated using a challenging dataset captured by camera mounted on unmanned air vehicle. The obtained results show that the proposed method achieves best results compared to other current state-of-the-art relevant methods.

Automatic license plate recognition (ALPR) is the extraction of vehicle license plate information from an image or a sequence of images. The extracted information can be used with or without a database in many applications, such as electronic payment systems (toll payment, parking fee payment), and freeway and arterial monitoring systems for traffic surveillance. The ALPR uses either a color, black and white, or infrared camera to take images. The quality of the acquired images is a major factor in the success of the ALPR. ALPR as a real-life application has to quickly and successfully process license plates under different environmental conditions, such as indoors, outdoors, day or night time. It should also be generalized to process license plates from different nations, provinces, or states. These plates usually contain different colors, are written in different languages, and use different fonts; some plates may have a single color background and others have background images. The license plates can be partially occluded by dirt, lighting, and towing accessories on the car. In this paper, we present a comprehensive review of the state-of-the-art techniques for ALPR. We categorize different ALPR techniques according to the features they used for each stage, and compare them in terms of pros, cons, recognition accuracy, and processing speed. Future forecasts of ALPR are given at the end.

Video-based automatic incident detection (AID) systems are increasingly being used in intelligent transportation systems (ITS). Video-based AID is a promising method of incident detection. However, the accuracy of video-based AID is heavily affected by environmental factors such as shadows, snow, rain, and glare. This paper presents a review of the different work done in the literature to detect outdoor environmental factors, namely, static shadows, snow, rain, and glare. Once these environmental conditions are detected, they can be compensated for, and hence, the accuracy of alarms detected by video-based AID systems will be enhanced. Based on the presented review, this paper will highlight potential research directions to address gaps that currently exist in detecting outdoor environmental conditions. This will lead to an overall enhancement in the reliability of video-based AID systems and, hence, pave the road for more usage of these systems in the future. Last, this paper suggests new contributions in the form of new suggested algorithmic ideas to detect environmental factors that affect AID systems accuracy.

Smart homes have enjoyed increasing popularity in recent years. In order for them to further expand their market share, users need to be able to fully control devices. Policies are one way for users to achieve such flexible control of devices. However, user policies often tend to interact in unwanted ways leading to unexpected behavior of devices. This paper describes the design of a run-time policy interaction management module (PIMM) that serves as manager for detecting and resolving interactions between user policies in KNX-based smart homes. This module extends the traditional KNX networking system with the ability to manage policy interactions. The module operates in the run-time S-mode of the KNX network and works as part of the engineering tool software (ETS) used to configure and control the operation of the KNX network in smart homes. The proposed module serves as the first of its kind that can be implemented inside the KNX networking system to detect and resolve unwanted policies interactions.

This paper deals with the problem of requirements interaction. We introduce a three level framework to detect requirements interactions at different levels of cost, time, and complexity. Level 2 where we use semiformal methods to detect interactions contains the main contribution of the research. Also we combine existing approaches (e.g. informal and formal) with our semiformal approach to provide a comprehensive framework for developers to use. The approach is illustrated using two case studies, one from the telecommunications domain and the other one being a lift control system. The results obtained are very encouraging with regards to the time and effort spent on requirements interaction detection.