Open-World Object Detection (OWOD) aims to detect known objects while identifying unknown instances. A fundamental challenge is the semantic ambiguity where unknown objects are absorbed into the background class during training, impairing their subsequent identification. To address this, we propose the Decoupled Evidential Detector (DEED), which is based on a novel decouple-and-unify strategy. DEED leverages Evidential Deep Learning (EDL), employing a Beta branch to exclusively model a ’pure’ background. This branch is then unified with the primary Dirichlet branch via a distillation mechanism, enabling the model to faithfully represent unknowns. Extensive experiments on standard OWOD benchmarks demonstrate that DEED significantly outperforms existing methods.

Shadow segmentation is sensitive to illumination shifts and scene diversity. Test-time adaptation provides a source-free solution, yet most methods assume large batches or stationary distributions, limiting online use. We present an online test-time adaptation framework for shadow segmentation, where data arrive sequentially in small batches without revisiting. Our method applies dual-stage filtering: image-level domain scoring with a frozen supervisor, followed by pixel-level soft masking guided by confidence, consistency, and structural cues. A brightness-guided module exploits illumination priors to refine shadow regions, while curriculum-based thresholding and exponential moving average teacher–student updates stabilize adaptation and reduce error accumulation. Experiments across six cross-domain settings on SBU, CUHK-Shadow, and ISTD demonstrate consistent gains over baselines in accuracy and boundary preservation, achieving robust generalization under diverse shifts.

As tennis continues to gain global popularity, there is a growing need for precise, scalable, and standardized training solutions. However, the evaluation and instruction of tennis techniques remain largely dependent on subjective assessments by experienced coaches, limiting consistency and automation. In this paper, we present TAGS (Tennis Action Guidance System), a unified framework that leverages large language models to generate expert-level evaluations and actionable feedback for tennis performance. Given a tennis video and a user-defined query, TAGS first extracts 3D skeletal keypoints, encodes the motion into a structured representation, and employs prompt-based LLMs to assess performance across five critical dimensions: stability, coordination, power generation, technical execution, and rhythm. To support this process, we introduce (1) a pose-language unified representation that aligns 3D motion data with natural language understanding, (2) structured prompt templates grounded in professional coaching expertise, and (3) a multi-dimensional evaluation protocol incorporating both human and automated assessments. In addition, we contribute the first large-scale dataset for tennis action evaluation, comprising 3,000 video-query-feedback triplets authored by certified coaches. Extensive experiments demonstrate that TAGS consistently outperforms existing baselines in delivering accurate, domain-specific, and actionable feedback. This work marks a step forward in developing interpretable, human-aligned systems for real-world sports action assessment and intelligent coaching.

Point cloud completion is crucial for downstream tasks in 3D visual perception. However, existing methods often struggle to generalize to real-world scans due to their heavy reliance on abundant paired point clouds for training and their neglect of the distribution shift between training and testing datasets. To address these limitations, this paper explores a practical and challenging setting: "source-free domain adaptive point cloud completion", where a well-trained source model must adapt to the target data distribution without access to source data, aiming to improve completion performance. To tackle this problem, we propose a novel method called "Dual-Stage Preservation and Fusion" (DSPF), which comprises two key training stages tailored to this new setting. In the source preservation stage, we introduce graph structural alignment and marginal feature alignment to preserve and transfer essential knowledge from the source domain. In the target fusion stage, we design a self-supervised loss to capture the geometric structure of target instances and establish a bidirectional interaction mechanism to transfer partial source knowledge to the target distribution. Extensive experiments on various cross-domain point cloud completion benchmarks demonstrate that our proposed DSPF significantly outperforms existing methods, validating its effectiveness and robustness in source-free domain adaptation scenarios. Our code is available at https://github.com/ZhiXia-SEU/DSPF.

Open-set domain adaptation (OSDA) considers a special domain adaptation problem in which the target domain contains novel categories that never appear in the well-labeled source domain. Unfortunately, prior efforts on OSDA simply detect and recognize all novel categories as one "unknown" group without further exploration. The demand for exploring these novel categories prompts us to consider the underlying multi-class structure and semantic description of those unknown categories in more detail. In this article, we propose a novel interpretable framework to accurately identify the seen categories in the target domain and effectively recover the semantic knowledge of the unseen categories with attributes and visual interpretations, which is referred to as Semantic Recovery Open-Set Domain Adaptation (SR-OSDA). Specifically, the proposed framework includes an explicit attribute explainable module and an implicit semantic interpretable module, which provide insight into the process of domain adaptation and the discovery of new categories. Furthermore, structure-preserving partial alignment is developed as a method of recognizing and aligning the visible categories across domains with the aid of domain-invariant feature learning. The visual-structural semantic attributes propagation is designed to provide smooth transitions from seen categories to unseen categories via visual-semantic mapping. Three new cross-domain SR-OSDA benchmarks are constructed in order to evaluate the proposed framework in novel and practical challenges. Experimental results and empirical analysis of our proposed solution to open-set recognition and semantic recovery demonstrate its superiority over other state-of-the-art solutions. Our source code is available at https://github.com/scottjingtt/XSROSDA.

Camera-LiDAR 3D object detection is currently becoming a crucial component in the field of autonomous driving perception. However, previous models only performed feature fusion in the deep-level BEV hierarchy when dealing with camera-LiDAR feature fusion. This approach lacks interaction with the shallow-level sensor features, which is beneficial in constructing the corresponding BEV features. However, a simple shallow-level feature interaction can introduce sensor noise caused by intrinsic and extrinsic camera calibration errors. To address this, we propose RoBiFusion, a novel camera-LiDAR 3D object detection framework designed for effective sensor feature interaction and mitigating sensor noise interference. This framework consists of three submodules: the Camera-LiDAR Feature Matching module, the LiDAR-to-Camera module, and the Camera-to-LiDAR module. Firstly, in the Camera-LiDAR Feature Matching module, we use the cross-attention module to dynamically match the camera features and the LiDAR features, which solves the problem of feature inconsistency caused by noise in the camera's intrinsic and extrinsic parameters. Secondly, in the LiDAR-to-Camera module, we propose a novel depth representation that can effectively mitigate LiDAR noise interference. Thirdly, in the Camera-to-LiDAR module, we introduce deformable attention to help LiDAR feature capture instance-level semantic features. Additionally, we design a novel differentiable and efficient grid sample module to accelerate the process since the bilinear grid sample module in deformable attention is time-consuming and not deployment-friendly. We compared RoBiFusion to the state-of-the-art BEVFusion on the nuScenes dataset and found that RoBiFusion surpasses BEVFusion by 1.5% mAP and 2.4% NDS. Furthermore, we designed a series of ablation experiments to verify the effectiveness of the aforementioned modules.

The creation of diverse and realistic driving scenarios has become essential to enhance perception and planning capabilities of the autonomous driving system. However, generating long-duration, surround-view consistent driving videos remains a significant challenge. To address this, we present UniMLVG, a unified framework designed to generate extended street multi-perspective videos under precise control. By integrating single- and multi-view driving videos into the training data, our approach updates a DiT-based diffusion model equipped with cross-frame and cross-view modules across three stages with multi training objectives, substantially boosting the diversity and quality of generated visual content. Importantly, we propose an innovative explicit viewpoint modeling approach for multi-view video generation to effectively improve motion transition consistency. Capable of handling various input reference formats (e.g., text, images, or video), our UniMLVG generates high-quality multi-view videos according to the corresponding condition constraints such as 3D bounding boxes or frame-level text descriptions. Compared to the best models with similar capabilities, our framework achieves improvements of 48.2% in FID and 35.2% in FVD.

Multi-source domain adaptation (MSDA) borrows intrinsic knowledge from well-annotated source domains to identify target visual signals. The main challenges are effectively mitigating cross-domain shift and extracting discriminative target features via the suitable source semantics. To overcome them, this paper proposes a novel Interpretable Prototype Network (IPNet) with channel-wise augmentation and multi-domain prototype mechanism. Specifically, IPNet explores the parameterized channel fusion paradigm across multiple source domains and target one to generate intermediate instances and achieve beneficial alignment. Moreover, IPNet analyzes contributions of source domains with interpretable learning approach and adjusts their effects on representations of target signals. Extensive experiments on three MSDA benchmark datasets suggest the advantages of our IPNet over others and exhibit the path of knowledge transfer.

This paper addresses the generalized motion style transfer problem, where the main challenge lies in the inability to apply discrete semantic labels for effective constraints. A cross-dataset style transfer framework is introduced to address key limitations in existing motion synthesis methods. Unlike prior approaches that rely on paired data or discrete semantic labels, our model transfers style through AdaIN-based adaptive instance normalization to encode content motion and style motion separately. A dual-path training strategy that combines style transfer generation and self-reconstruction ensures content preservation. Then, the Dual-Stream Denoising Network (DSDN) integrates LSTM processing for dynamic correction and spatio-temporal autoencoders for static motion refinement, with an Adaptive Gating Fusion Unit (AGFU) that dynamically balances stride amplitude and joint angles. Innovations include cross-dataset transfer without predefined labels, raw-sequence LSTM processing for motion correction, and adaptive fusion architecture outperforming single-method approaches. Validated on the HumanML3D and Mocap datasets, our framework demonstrates significantly enhanced cross-dataset style transfer efficacy.

In domain adaption (DA), joint maximum mean discrepancy (JMMD), as a famous distribution-distance metric, aims to measure joint probability distribution difference between the source domain and target domain, while it is still not fully explored and especially hard to be applied into a subspace-learning framework as its empirical estimation involves a tensor-product operator whose partial derivative is difficult to obtain. To solve this issue, we deduce a concise JMMD based on the Representer theorem that avoids the tensor-product operator and obtains two essential findings. First, we reveal the uniformity of JMMD by proving that previous marginal, class conditional, and weighted class conditional probability distribution distances are three special cases of JMMD with different label reproducing kernels. Second, inspired by graph embedding, we observe that the similarity weights, which strengthen the intra-class compactness in the graph of Hilbert Schmidt independence criterion (HSIC), take opposite signs in the graph of JMMD, revealing why JMMD degrades the feature discrimination. This motivates us to propose a novel loss JMMD-HSIC by jointly considering JMMD and HSIC to promote discrimination of JMMD. Extensive experiments on several cross-domain datasets could demonstrate the validity of our revealed theoretical results and the effectiveness of our proposed JMMD-HSIC.

In the context of smart agriculture, effective monitoring of crop growth requires a 3D point cloud compression method that preserves detailed features of the primary plant structures while suppressing less relevant background information. However, achieving this balance is challenging due to the uniform treatment of spatial regions in conventional compression schemes. To address this need, we propose a resource-efficient importance-aware adaptive compression framework for point clouds. Specifically, after performing semantic segmentation on the raw point cloud, we quantify the semantic importance of each segmented region and map the importance scores to adaptive compression ratios. These ratios are then embedded into the latent representation during encoding and parsed during decoding to guide the reconstruction precision. This enables a more efficient allocation of bits to semantically critical regions. We validate the effectiveness of the proposed method on both our self-collected crop dataset and the publicly available Pheno4D dataset, demonstrating its feasibility and improved performance in balancing compression rate and reconstruction accuracy.

Source-free domain adaptation (SFDA) aims to transfer knowledge from the well-trained source model and optimize it to adapt target data distribution. SFDA methods are suitable for medical image segmentation task due to its data-privacy protection and achieve promising performances. However, cross-domain distribution shift makes it difficult for the adapted model to provide accurate decisions on several hard instances and negatively affects model generalization. To overcome this limitation, a novel method `supportive negatives spectral augmentation' (SNSA) is presented in this work. Concretely, SNSA includes the instance selection mechanism to automatically discover a few hard samples for which source model produces incorrect predictions. And, active learning strategy is adopted to re-calibrate their predictive masks. Moreover, SNSA deploys the spectral augmentation between hard instances and others to encourage source model to gradually capture and adapt the attributions of target distribution. Considerable experimental studies demonstrate that annotating merely 4%~5% of negative instances from the target domain significantly improves segmentation performance over previous methods.

Human action video recognition has recently attracted more attention in applications such as video security and sports posture correction. Popular solutions, including graph convolutional networks (GCNs) that model the human skeleton as a spatiotemporal graph, have proven very effective. GCNs-based methods with stacked blocks usually utilize top-layer semantics for classification/annotation purposes. Although the global features learned through the procedure are suitable for the general classification, they have difficulty capturing fine-grained action change across adjacent frames - decisive factors in sports actions. In this paper, we propose a novel “Cross-block Fine-grained Semantic Cascade (CFSC)” module to overcome this challenge. In summary, the proposed CFSC progressively integrates shallow visual knowledge into high-level blocks to allow networks to focus on action details. In particular, the CFSC module utilizes the GCN feature maps produced at different levels, as well as aggregated features from proceeding levels to consolidate fine-grained features. In addition, a dedicated temporal convolution is applied at each level to learn short-term temporal features, which will be carried over from shallow to deep layers to maximize the leverage of low-level details. This cross-block feature aggregation methodology, capable of mitigating the loss of fine-grained information, has resulted in improved performance. Last, FD-7, a new action recognition dataset for fencing sports, was collected and will be made publicly available. Experimental results and empirical analysis on public benchmarks (FSD-10) and self-collected (FD-7) demonstrate the advantage of our CFSC module on learning discriminative patterns for action classification over others.

Source-free domain adaptation (SFDA) assumes that model adaptation only accesses the well-learned source model and unlabeled target instances for knowledge transfer. However, cross-domain distribution shift easily triggers invalid discriminative semantics from source model on recognizing the target samples. Hence, understanding the specific content of discriminative pattern and adjusting their representation in target domain become the important key to overcome SFDA. To achieve such a vision, this paper proposes a novel explanation paradigm "Discriminative Pattern Calibration (DPC)" mechanism on solving SFDA issue. Concretely, DPC first utilizes learning network to infer the discriminative regions on the target images and specifically emphasizes them in feature space to enhance their representation. Moreover, DPC relies on the attention-reversed mixup mechanism to augment more samples and improve the robustness of the classifier. Considerable experimental results and studies suggest that the effectiveness of our DPC in enhancing the performance of existing SFDA baselines.

Deep neural networks have been successfully applied in many computer vision tasks. However, these models suffer catastrophic forgetting when learning new knowledge incrementally. To overcome the stability-plasticity dilemma, class incremental learning (CIL) has been widely discussed recently. The state-of-the-art CIL methods mainly leverage additional exemplar sets, thus memory costly and may raise privacy issues. To that end, we propose an autonomous generative feature replay (AGFR) framework without using exemplar sets. It consists of three modules: the feature extractor module, the feature generator module, and the unified classification module. First, to stabilize features over tasks, robust feature extractors are learned in a self-supervised manner and thus generalize well to unseen data. Second, instead of using exemplar sets or producing raw images, we propose an autonomous generative feature replay scheme to constantly update unified classifier in CIL without saving any image data. This strategy avoids overwhelming memory usage or poor quality of the generated raw images. Experiments demonstrate that our method achieves state-of-the-art performance in terms of average classification accuracy. ⋆

Speech-driven 3D facial animation aims to synthesize 3D talking head animations with precise lip movements and rich stylistic expressions. However, existing methods exhibit two limitations: 1) they mostly focused on emotionless facial animation modeling, neglecting the importance of emotional expression, due to the lack of high-quality 3D emotional talking head datasets, and 2) several latest works treated emotional intensity as a global controllable parameter, akin to emotional or speaker style, leading to over-smoothed emotional expressions in their outcomes. To address these challenges, we first collect a 3D talking head dataset comprising five emotional styles with a set of coefficients based on the MetaHuman character model and then propose an end-to-end deep neural network, DEITalk, which conditions on speech and emotional style labels to generate realistic facial animation with dynamic expressions. To model emotional saliency variations in long-term audio contexts, we design a dynamic emotional intensity (DEI) modeling module and a dynamic positional encoding (DPE) strategy. The former extracts implicit representations of emotional intensity from speech features and utilizes them as local (high temporal frequency) emotional supervision, whereas the latter offers abilities to generalize to longer speech sequences. Moreover, we introduce an emotion-guided feature fusion decoder and a four-way loss function to generate emotion-enhanced 3D facial animation with controllable emotional styles. Extensive experimental results demonstrate that our method outperforms existing state-of-the-art methods. Our video demo and dataset are available at https://github.com/KangShen-seu/DEITalk.

Speech-driven 3D facial animation technology has been developed for years, but its practical application still lacks expectations. The main challenges lie in data limitations, lip alignment, and the naturalness of facial expressions. Although lip alignment has seen many related studies, existing methods struggle to synthesize natural and realistic expressions, resulting in a mechanical and stiff appearance of facial animations. Even with some research extracting emotional features from speech, the randomness of facial movements limits the effective expression of emotions. To address this issue, this paper proposes a method called CSTalk (Correlation Supervised) that models the correlations among different regions of facial movements and supervises the training of the generative model to generate realistic expressions that conform to human facial motion patterns. To generate more intricate animations, we employ a rich set of control parameters based on the metahuman character model and capture a dataset for five different emotions. We train a generative network using an autoencoder structure and input an emotion embedding vector to achieve the generation of user-control expressions. Experimental results demonstrate that our method outperforms existing state-of-the-art methods.

Deep clustering as an important branch of unsupervised representation learning focuses on embedding semantically similar samples into the identical feature space. This core demand inspires the exploration of contrastive learning and subspace clustering. However, these solutions always rely on the basic assumption that there are sufficient and category-balanced samples for generating valid high-level representation. This hypothesis actually is too strict to be satisfied for real-world applications. To overcome such a challenge, the natural strategy is utilizing generative models to augment considerable instances. How to use these novel samples to effectively fulfill clustering performance improvement is still difficult and under-explored. In this paper, we propose a novel Generative Calibration Clustering (GCC) method to delicately incorporate feature learning and augmentation into clustering procedure. First, we develop a discriminative feature alignment mechanism to discover intrinsic relationship across real and generated samples. Second, we design a self-supervised metric learning to generate more reliable cluster assignment to boost the conditional diffusion generation. Extensive experimental results on three benchmarks validate the effectiveness and advantage of our proposed method over the state-of-the-art methods.

Domain adaptation (DA) has recently drawn a lot of attention, as it facilitates unlabeled target learning by borrowing knowledge from an external source domain. Most existing DA solutions seek to align feature representations between the labeled source and unlabeled target data. However, the scarcity of target data easily results in negative transfer, as it misleads the cross DA to the dominance of the source. To address the challenging few-shot domain adaptation (FSDA) problem, in this article, we propose a novel marginalized augmented FSDA (MAF) approach to address the cross-domain distribution disparity and insufficiency of target data simultaneously. On the one hand, cross-domain continuity augmentation (CCA) synthesizes abundant intermediate patterns across domains leading to a continuous domain-invariant latent space. On the other hand, sufficient source-supervised semantic augmentation (SSA) is explored to progressively diversify the conditional distribution within and across domains. Moreover, the proposed augmentation strategies are implemented efficiently via an expected transferable cross-entropy (CE) loss over the augmented distribution instead of explicit data synthesis, and minimizing the upper bound of the expected loss introduces negligible extra computing cost. Experimentally, our method outperforms the state of the art in various FSDA benchmarks, which demonstrates the effectiveness and contribution of our work. Our source code is provided at https://github.com/scottjingtt/MAF.git.