Safety-critical mental-health support systems must distinguish when supportive conversation is appropriate from when free-form generation should be blocked. This paper presents Anian, a safety-gated multimodal AI backend for perinatal mental-health support and mindfulness-intervention routing. Anian is not intended to diagnose psychiatric conditions or replace clinical care or crisis intervention. Its modular pipeline places generative AI downstream of structured state representation, conservative risk fusion, and response gating. User text or voice-derived ASR transcripts are mapped into four linked layers: L1 emotion states, L2 psychosocial constructs, L3 safety risk, and L4 intervention routes. Local text- and rule-based safety evidence is fused with external voice-derived evidence using a highest-risk-priority rule, S_fusion = max(S_local, S_external). At moderate or high fused risk, ordinary AI-generated responses and text-to-speech delivery are blocked and replaced by fixed safety content and prompts for human support. An internal prototype evaluation used approximately 858,295 normalized records from public emotion, dialogue, mental-health-related, and Chinese dialogue corpora within a weak-label and rule-derived framework. Micro-F1 scores were 0.9604 for L1 emotion classification, 0.9144 for L2 psychosocial constructs, and 0.9742 for L4 routing. In a controlled safety stress test of 233 samples, the L3 rule engine achieved high-risk recall of 1.0000 within predefined scenarios. These findings support the internal feasibility of the label framework and gating logic but do not establish clinical validity, diagnostic accuracy, real-world safety, or effectiveness. We report the architecture, ontology, safety-fusion mechanism, prototype evaluation, error-analysis plan, and roadmap for expert-reviewed and real-world validation.

Unsupervised domain adaptive semantic segmentation aims to transfer knowledge from the annotated source domain to the unlabeled target domain. Recently, self-training methods have gained substantial attention, which leverage high-confidence predictions in the target domain as pseudo labels for supervision. However, limited exploration of intra-class variations across domains, including significant visual differences within each category, has led to misalignment between feature distribution across domains. In this article, we present a unified non-parametric distance-based online clustering method to efficiently maintain multiple centroid-based prototypes within each category subspace instead of one prototype for each category subspace, which enables prototypes to possess the capacity for richer feature representation. Then, considering the variance across different dimensions of a feature representation, we then extend the prototypes from centroid-based ones to distribution-based ones. Specifically, each subspace is modeled using a Gaussian mixture model which includes several anisotropic Gaussian distributions, aimed at prioritizing discriminative dimensions and obtaining a finer measurement of the pixel-to-prototype similarity. Meanwhile, a category-aware feature space is achieved through pixel-to-prototype contrastive learning to ensure the compactness of pixel features in the same subcategory and drive the separation between pixel features of different subcategories. What's more, multi-resolution features are utilized to promote diversity and robustness among intra-class prototypes. Experiments validate the competitiveness of our two prototype-based methods against existing state-of-the-art methods, with a mIoU of 76.8% on GTA -> Cityscapes, 68.4% on Synthia -> Cityscapes, 54.5% on Cityscapes -> DarkZurich and 56.4% on Cityscapes -> ACDC. Notably, our method is able to seamlessly integrate with existing UDA methods.

Deep learning has advanced medical image segmentation, especially for optic disc (OD) and cup detection. While convolutional neural networks (CNNs) struggle with long-range dependencies, Transformer-based architectures have emerged to address this limitation.However, complete replacement of CNNs with Transformers may impair local feature extraction. Additionally, reliance on expert-annotated datasets makes supervised learning costly. To overcome these limitations, we present DAK-Former, a novel self-supervised contrastive learning approach for OD and cup segmentation. Our approach introduces a new attention mechanism that combines K-nearest neighbors (KNN) with Transformer components, along with a deformable aggregation attention module to improve global feature representation. Additionally, our method uses multiple types of medical imaging data, including MRI, CT scans, and X-rays. Using contrastive learning, we match encoded queries with a dictionary of encoded keys, allowing the network to learn meaningful unsupervised feature representations. We evaluated DAK-Former on two publicly available fundus image datasets and compared it with state-of-the-art methods. Experimental results show that DAK-Former is highly effective and consistently outperforms existing approaches.

As a focal point of research in various fields, human body language understanding has long been a subject of intense interest. Within this realm, the exploration of emotion recognition through the analysis of facial expressions, voice patterns, and physiological signals, holds significant practical value. Compared with unimodal approaches, multimodal emotion recognition models leverage complementary information from vision, acoustic, and language modalities to robust perceive the human sentiment attitudes. However, the heterogeneity among modality signals leads to significant domain shifts, posing challenges for achieving balanced fusion. In this paper, we propose a Domain-separated Bottleneck Attention fusion framework (DBA Framework) for human multimodal emotion recognition with lower computational complexity. Specifically, we partition each modality into two distinct domains: the invariant/private domain. The invariant domain contains crucial shared information, while the private domain aims to capture modality-specific representations. For the decomposed features, we introduce two sets of bottleneck cross-attention modules to effectively utilize the complementarity between domains to reduce redundant information. In each module, we interweave two Fusion Adapter blocks into the Self-Attention Transformer backbone. Each Fusion Adapter block integrates a small group of latent tokens as bridges for inter-modal and inter-domain interactions, mitigating the adverse effects of modality distribution differences and lowering computational costs. Extensive experimental results demonstrate that our method outperforms state-of-the-art (SOTA) approaches across three widely used benchmark datasets.

This paper presents our championship-winning method for valence-arousal (VA) estimation in the 8th Affective Behavior Analysis in-the-Wild (ABAW) competition. Our approach skillfully integrates visual and audio information through a sophisticated multimodal framework. The visual branch leverages a pre-trained ResNet model to extract spatial features from facial images. The audio branches utilize pre-trained VGG models to extract both VGGish and LogMel features from speech signals. These extracted features undergo advanced temporal modeling using specifically designed Temporal Convolutional Networks (TCNs). Following this, we employ sophisticated cross-modal attention mechanisms, enabling visual features to interact with audio features through meticulously structured query-keyvalue attention processes. Finally, the processed features are concatenated and passed through a meticulously designed regression layer to accurately predict valence and arousal. Our method demonstrates outstanding performance on the Aff-Wild2 dataset, significantly advancing the field of effective multimodal fusion for VA estimation in unconstrained environments. By achieving first place in the ABAW VA track, our approach establishes a new benchmark in this challenging domain, demonstrating superior accuracy and robustness in real-world emotion recognition scenarios.

Facial expression recognition is a challenging classification task that holds broad application prospects in the field of human-computer interaction. This paper aims to introduce the method we will adopt in the 8th Affective and Behavioral Analysis in the Wild (ABAW) Competition, which will be held during the Conference on Computer Vision and Pattern Recognition (CVPR) in 2025. First of all, we apply the frequency masking technique and the method of extracting data at equal time intervals to conduct targeted processing on the original videos. Then, based on the residual hybrid convolutional neural network and the multi-branch convolutional neural network respectively, we design feature extraction models for image and audio sequences. In particular, we propose a global channel-spatial attention mechanism to enhance the features initially extracted from both the audio and image modalities respectively. Finally, we adopt a decision fusion strategy based on the proportional criterion to fuse the classification results of the two single modalities, obtain an emotion probability vector, and output the final emotional classification. We also design a coarse - fine granularity loss function to optimize the performance of the entire network, which effectively improves the accuracy of facial expression recognition. In the facial expression recognition task of the 8th ABAW Competition, our method ranked third on the official validation set. This result fully confirms the effectiveness and competitiveness of the method we have proposed.

Facial expression recognition is a challenging classification task that holds broad application prospects in the field of human-computer interaction. This paper aims to introduce the method we will adopt in the 8th Affective and Behavioral Analysis in the Wild (ABAW) Competition, which will be held during the Conference on Computer Vision and Pattern Recognition (CVPR) in 2025.First of all, we apply the frequency masking technique and the method of extracting data at equal time intervals to conduct targeted processing on the original videos. Then, based on the residual hybrid convolutional neural network and the multi-branch convolutional neural network respectively, we design feature extraction models for image and audio sequences. In particular, we propose a global channel-spatial attention mechanism to enhance the features initially extracted from both the audio and image modalities respectively.Finally, we adopt a decision fusion strategy based on the proportional criterion to fuse the classification results of the two single modalities, obtain an emotion probability vector, and output the final emotional classification. We also design a coarse - fine granularity loss function to optimize the performance of the entire network, which effectively improves the accuracy of facial expression recognition.In the facial expression recognition task of the 8th ABAW Competition, our method ranked third on the official validation set. This result fully confirms the effectiveness and competitiveness of the method we have proposed.

Abstract In the infrared imaging scene where the target is at a long distance and the background is cluttered, due to the interference of noise and background texture information, the infrared image is prone to problems such as low contrast between the target and the background, and feature confusion, which makes it difficult to accurately extract and detect the target. To solve this problem, firstly, the infrared image is enhanced by combining DDE and MSR algorithm to improve the contrast and detail visibility of the image. For the RT-DETR network structure, the EMA attention mechanism is introduced into the backbone to enhance the feature extraction ability of the model by extracting context information. The CAMixing convolutional attention module is introduced into CCFM, and the multi-scale convolutional self-attention mechanism is introduced to focus on local information and enhance the detection ability of small targets. The filtering rules of the prediction box are improved, combined with Shape-IoU, and the convergence speed of the loss function in the detection and the detection accuracy of small targets are improved by paying attention to the influence of the intrinsic properties of the bounding box itself on the regression. In the experiment, the infrared weak target image dataset of the National University of Defense Technology was selected, labeled and trained. Experimental results show that compared with the original DETR algorithm, the average precision of the improved algorithm (mAP) is increased by 3.2%, and it can effectively detect infrared weak and small targets in different complex backgrounds, which reflects good robustness and adaptability, and can be effectively applied to infrared weak and small target detection in complex backgrounds.

Machine learning models are increasingly recognized as valuable intellectual property (IP), prompting the development of a range of watermarking techniques aimed at safeguarding the IP of these models. However, in the context of federated learning (FL) models involving multiple owners, such as the participants in FL model training, conventional techniques designed for single-owner models prove ineffective due to limitations in their capacity and robustness. Few work has explored how to effectively embed watermarks to FL models for multiple-owners, which is non-trivial, especially when the number of owners is large. To fill this gap, we first analyze the capacity of existing watermarking methods. Second, we propose FedMark, a general large-capacity watermarking mechanism for FL, which leverages the Bloom Filter to achieve conflict-free watermarking of a large number of participants. Moreover, we propose a secret-sharing-based verification method to improve the watermarking robustness against false positives caused by Bloom Filter. Finally, comprehensive experiments show that our design can support over 150 participants to embed watermarks while the model accuracy varies within 1%, and is robust to non-independent identical distributed data, different participant selection rates, model modifications, permutation attacks, scaling attacks and forging attacks.

Deep networks have achieved great success in forest fire detection by exploiting visible light images. However, visible light images are susceptible to strong light, smoke, and obstruction interference. The infrared image has high sensitivity to temperature changes of targets, which can alleviate the deficiency of visible light image. Due to the significant distribution shift between visible light and infrared images, directly using the visible light-based pre-trained network for infrared forest fire results in a significant decrease in performance. To resolve this issue, this paper proposes an infrared image forest fire detection system based on domain adaptive learning. We adopt two YOLOv5 frameworks to extract features from visible light images (source domain) and infrared images (target domain). To align the features of the two domains, we construct a novel adaptation learning mechanism based on Kullback-Leibler (KL) loss and feature maximum mean discrepancy (FMMD) loss. We conducted extensive comparative experiments on two publicly available datasets to verify the effectiveness of the proposed model. All experimental results indicate that our proposed domain adaptive learning mechanism effectively improves the performance of infrared forest fire detection.

This paper focuses on the advancement of single object tracking technologies in computer vision, which have broad applications including robotic vision, video surveillance, and sports video analysis. Current methods relying solely on the target's initial visual information encounter performance bottlenecks and limited applications, due to the scarcity of target semantics in appearance features and the continuous change in the target's appearance. To address these issues, we propose a novel approach, combining visual-language dual-modal single object tracking, that leverages natural language descriptions to enrich the semantic information of the moving target. We introduce a dual-modal single-object tracking algorithm based on local correspondence modeling. The algorithm decomposes visual features into multiple local visual semantic features and pairs them with local language features extracted from natural language descriptions. In addition, we also propose a new global relocalization method that utilizes visual language bimodal information to perceive target disappearance and misalignment and adaptively reposition the target in the entire image. This improves the tracker's ability to adapt to changes in target appearance over long periods of time, enabling long-term single target tracking based on bimodal semantic and motion information. Experimental results show that our model outperforms stateof-the-art methods, which demonstrates the effectiveness and efficiency of our approach

In a conversation, it is crucial for the listener to provide appropriate reactions to the speaker, as the dialogue becomes challenging to sustain without the listener's involvement. Consequently, responsive listening head synthesis has become an important task. However, the existing methods fail to adequately utilize the audio and video of the speaker to generate listening heads, resulting in unnatural or even distorted generated videos. In this paper, we propose a framework to effectively encode audio and video features to address this problem. The framework includes speaker audio encoder, speaker video encoder, a Dual-Stream Prediction Network and a rendering network. The audio encoder utilizes a transformer encoder to encode the audio features, enabling better focus on the contextual features of long audio inputs. The speaker video encoder uses a 3D morphable model (3DMM) to extract the speaker's video features. Then we fuse the speaker's audio-video features with the listener's identity information extracted by 3DMM to preserve the listener's style. Additionally, a Dual-Stream Prediction Network is introduced to further enhance the prediction capability of the network. Finally, a rendering network is used to generate the listening heads based on the result of the prediction network. Comprehensive experiments demonstrate that our approach is capable of generating responsive listening heads with higher visual quality, better naturalness and higher reconstruction fidelity.

The goal of unsupervised domain adaptive semantic segmentation (UDA-SS) is to learn a model using annotated data from the source domain and generate accurate dense predictions for the unlabeled target domain. UDA methods based on Transformer utilize self-attention mechanism to learn features within source and target domains. However, in the presence of significant distribution shift between the two domains, the noisy pseudo-labels could hinder the model’s adaptation to the target domain. In this work, we proposed to incorporate self-attention and cross-domain attention to learn domain-invariant features. Specifically, we design a weight-sharing multi-branch cross-domain Transformer, where the cross-domain branch is used to align domains at the feature level with the aid of cross-domain attention. Moreover, we introduce an adaptive thresholding strategy for pseudo-label selection, which dynamically adjusts the proportion of pseudo-labels that are used in training based on the model’s adaptation status. Our approach guarantees the reliability of the pseudo labels while allowing more target domain samples to contribute to model training. Extensive experiments show that our proposed method consistently outperforms the baseline and achieves competitive results on GTA5 → Cityscapes, Synthia → Cityscapes, and Cityscapes → ACDC benchmark.

Human facial expressions can convey a great deal of information in daily life. Spotting macro-expression (MaE) and micro-expression (ME) intervals from long video sequences is a difficult challenge. In this paper, we propose an efficient framework for the expression spotting task. This framework consists of three main modules: Face Cropping and Alignment Module (FCAM), optical flow Feature Extraction Module (FEM), and expression Proposal Generation Module (PGM). The noise of optical flow features is reduced by face cropping and alignment, and the Main Directional Mean Optical Flow Feature of the regions of interest is extracted as the feature for expression spotting. Finally, the expression intervals are spotted by our designed expression proposal generation module. Our approach achieves very good results on the SAMM Long Videos and CAS(ME)^2. To demonstrate the transferability of our method, we tested it on the MEGC2023 unseen dataset and finally achieved the third place, proving the effectiveness of our method.

卫星能量与存储有限,对密集点目标进行观测时,存在观测任务间互斥、需要观测较多次数的问题.针对成像卫星密集点目标任务规划,本文改进了点目标聚类策略,使用改进的最大度团划分方法生成尽量少数量的任务团,同时使获取的任务团优先级之和更大;建立了考虑时间、能量、观测角度等约束的满足模型,为获得更高的观测总收益,并减少侧摆带来的能量损耗,以观测任务优先级之和与平均侧摆角度为优化目标,设计一种考虑基于侧摆优化的启发式蚁群算法对模型进行求解,最后,用仿真实验验证所提出算法的可行性和高效性,为卫星密集点目标任务规划提供技术支持.

Instance segmentation is a fundamental computer vision task with widespread applications. Numerous novel methods have been proposed to address this task. However, limited data and occlusion are common issues that hinder the practical application of instance segmentation. In this paper, we address limited data issue by employing image-level data augmentation. Additionally, to address the occlusion issue, we propose Balanced Occlusion Aware Copy-Paste (BOACP), a method that can not only increase the number of instances in images but also balance occluded instances at the image level. This method can enhance the performance of model on occluded instances. For the model, we utilize the Hybrid Task Cascade (HTC) based on CBSwin-Base and CBFPN. Moreover, we conduct additional experiments to explore the Occlusion Metric (OM). Experimental results demonstrate the effectiveness of our proposed approach, and we achieve the first place in the first phase of DeepSportRadar Instance Segmentation Challenge in ACM MMSports 2023 Workshop.

Cricket Bowl Release Detection aims to segment specific portions of bowl release actions occurring in multiple videos, with a focus on detecting the entire time window of this action. Unlike traditional detection tasks that identify action categories at a specific moment, this task involves identifying events that typically span around 100 frames and require recognizing all instances of the bowl release action in the video. Strictly speaking, this task falls under a branch of temporal action detection. With the advancement of deep neural networks, recent works have proposed deep learning-based approaches to address this task. However, due to the challenge of unclear action boundaries in videos, many existing methods perform poorly on the DeepSportradar Cricket Bowl Release Dataset. To more accurately identify specific portions of the bowl release action in videos, we adopt a one-stage architecture based on Relative Boundary Modeling. Specifically, our method consists of three stages. In the first stage, we use the Inflated 3D ConvNet (I3D) model to extract spatio-temporal features from the input videos. In the second stage, we utilize Temporal Action Detection with Relative Boundary Modeling (TriDet) to model the boundaries of the bowl release action's specific portions based on the relative relationships between different time moments, thereby predicting the action's time window. Lastly, as the target events typically span around 100 frames and the predicted time windows may exhibit overlapping regions based on confidence scores, we implement a post-processing step to merge and filter these outputs, resulting in the final submission results. We conducted extensive experiments to demonstrate that our proposed method achieves superior performance. Additionally, we evaluated the training techniques of existing approaches. Our proposed method achieves a PQ score of 0.519, an SQ score of 0.822, and an RQ score of 0.632 on the challenge set of the DeepSportradar Cricket Bowl Release Dataset. Through this approach, our team, USTC\_IAT\_United, won the third place in the first phase of the DeepSportradar Cricket Bowl Release Challenge.

Person Re-Identification (Re-ID) emerges as a important technique in sports analytics, enabling the accurate matching and recognition of players throughout a game. The fundamental objective of the Person Re-ID task is to identify the same player across diverse camera views, thus establishing their identity association over time. Generally speaking, the difficulty of the person Re-Identification (Re-ID) task lies in the perspective changes, occlusion phenomena, and posture changes caused by different camera placements and angles. In particular, for the Synergy re-identification dataset, the overlapping occlusion phenomenon between players and the low resolution and motion blur of the image make the Re-ID task challenging. In this paper, we analyze the impact of different data augmentations on this dataset and find effective augmentation methods. Meanwhile, we adopt a contrastive image-to-image training method and achieve higher results with the class-independent InfoNCE loss. We also quantitatively compare it with class-related ID loss. Finally, we employ the k-reciprocal re-ranking method to reorganize and optimize the distance matrix output by the model. The above process enables a single model to have good retrieval performance on the Synergy re-identification dataset. Then, we summarize the key factors affecting the retrieval performance on this dataset. To further improve the retrieval performance, we propose an efficient model/rank fusion method to fuse the retrieval results of different models from two perspectives of similarity and dissimilarity. Our proposed method achieves 98.81% mAP on the challenge set of the Synergy re-identification dataset, with which our team achieved 1st place in the DeepSportRadar player re-identification challenge 2023.

计算机程序设计是众多高校的公共基础课程,教学目标要求理论与实践并重.调研发现,学生的计算机基础差距很大,如何在统一的教学大纲和考核要求下,针对差异化的个体开展实验教学工作,是一项具有挑战性的难题.文章从学生能力培养的角度出发,采用现代教育技术,针对不同起点的学生,基于本地、服务端、云端等平台建立了一套多维度的实验教学体系.教学实践表明,该体系具有教学对象适用范围广、各平台互补性强的特点,能实现因材施教的目标,为计算机基础课程的实验教学提供一种可借鉴的方案.

Identifying abnormal behaviors in the crowd and taking corresponding intervention measures in time can prevent the occurrence of dangerous events. Due to this problem, we propose the crowd behavior intervention model based on emotional contagion, and use dynamic ways of emotional intervention to intervene group behavior. In the model, crowd behavior recognition module is introduced into crowd behavior intervention, which obtains the state of crowd behavior to support intervention decisions, improving the timeliness of interventions. Based on the identified behavior state of crowd, the problem of optimising the intervention strategy is transformed into the problem of maximising the ratio of the benefits to the costs of the intervention, when we use Intervention Optimization-Genetic Algorithm(IO-GA) to improve the efficiency of interventions. Experiments on the Motion Emotion Dataset(MED) demonstrate the effectiveness, timeliness and superiority of the crowd behavior intervention model.