Segmenting small, low-contrast anatomical structures and classifying their pathological status in ultrasound (US) images remain challenging tasks in computer vision, especially under the noise and ambiguity inherent in real-world clinical data. Papillary thyroid microcarcinoma (PTMC), characterized by nodules <= 1.0 cm, exemplifies these challenges where both precise segmentation and accurate lymph node metastasis (LNM) prediction are essential for informed clinical decisions. We propose SynTaskNet, a synergistic multi-task learning (MTL) architecture that jointly performs PTMC nodule segmentation and LNM classification from US images. Built upon a DenseNet201 backbone, SynTaskNet incorporates several specialized modules: a Coordinated Depth-wise Convolution (CDC) layer for enhancing spatial features, an Adaptive Context Block (ACB) for embedding contextual dependencies, and a Multi-scale Contextual Boundary Attention (MCBA) module to improve boundary localization in low-contrast regions. To strengthen task interaction, we introduce a Selective Enhancement Fusion (SEF) mechanism that hierarchically integrates features across three semantic levels, enabling effective information exchange between segmentation and classification branches. On top of this, we formulate a synergistic learning scheme wherein an Auxiliary Segmentation Map (ASM) generated by the segmentation decoder is injected into SEF's third class-specific fusion path to guide LNM classification. In parallel, the predicted LNM label is concatenated with the third-path SEF output to refine the Final Segmentation Map (FSM), enabling bidirectional task reinforcement. Extensive evaluations on a dedicated PTMC US dataset demonstrate that SynTaskNet achieves state-of-the-art performance, with a Dice score of 93.0% for segmentation and a classification accuracy of 94.2% for LNM prediction, validating its clinical relevance and technical efficacy.

Federated learning (FL) enables multiple clients to learn collaboratively in a distributed way, allowing for privacy protection. However, the real-world nonindependent and identically distributed (non-IID) data will lead to client drift, which degrades the performance of FL. Interestingly, we find that the logit difference between the local and global models increases as the model is continuously updated, which is the primary factor behind performance degradation. This is mainly due to catastrophic forgetting caused by non-IID data between clients. To alleviate this problem, we propose a new algorithm, named FedCSD, a class-prototype similarity distillation in a federated framework to align the logits of local and global models. FedCSD does not simply transfer global knowledge to local clients, as an insufficiently trained global model cannot provide reliable knowledge, i.e., class similarity information, and its wrong soft labels will mislead the optimization of local models. Concretely, FedCSD leverages the similarity between local logits and the global prototype to refine the global logits, thereby enhancing its class similarity information. Furthermore, FedCSD adopts an adaptive mask to filter out the terrible soft labels of the global models, thereby preventing them from misleading local optimization. Extensive experiments demonstrate the superiority of our method over the state-of-the-art FL approaches in various non-IID settings. Code is publicly available at https://github.com/IAMJackYan/FedCSD

Generating full-body humans in 360$^{\circ }$∘ has broad applications in digital entertainment, online education and art design. Existing works primarily rely on coarse conditions such as body pose to guide the generation, lacking detailed control over the synthesized results. Regarding this limitation, sketches offer a promising alternative as an expressive condition that enables more explicit and precise control. However, current sketch-based generation methods focus on faces or common objects, how to transfer sketches into 360 $^{\circ }$∘ full-body humans remains unexplored. To bridge this gap, we propose Sketch2Avatar, the first generative model to achieve 3D full-body human generation from hand-drawn sketches. Our model is capable of synthesizing sketch-aligned and 360$^{\circ }$∘-consistent full-body human images by leveraging the geometry information extracted from sketches to guide the 3D representation generation and neural rendering. Specifically, we propose sketchguided 3D representation generation to model the 3D human and maintain the alignment between input sketches and generated humans. Our transformer-based generator incorporates spatial feature guidance and latent modulation derived from sketches to produce high-quality 3D representations. Additionally, our designed bodyaware neural rendering utilizes 3D human body priors from sketches, simplifying the learning of articulated body poses and complex body shapes. To train and evaluate our model, we construct a large-scale dataset comprising approximately 19 K 2D full-body human images and their corresponding sketches in a hand-drawn style. Experimental results demonstrate that our Sketch2Avatar can transfer hand-drawn sketches into photo-realistic 360$^{\circ }$∘ full-body human images with precise sketch-human alignment. Ablation studies further validate the effectiveness of our design choices.

Existing clothing image editing methods typically rely on single-modal traditional models. However, as user demands evolve, these methods still face challenges in global consistency, fine-grained control, and intuitive interaction. Traditional GAN-based approaches produce limited editing effects, while current diffusion-based methods struggle with region-aware editing. To address these issues, we propose MIGEDIT, a multimodal interactive clothing editing process framework. It integrates potential spatial optimization, inversion guidance, and region aware editing into a pre trained diffusion model. MIGEDIT supports the generation of clothing images from clothing sketches and supports point based interactive editing as well as text or region guided modifications, enabling precise and flexible clothing adjustments. Experimental results demonstrate that MIGEDIT outperforms existing methods in visual quality and editing accuracy, making it well-suited for intelligent fashion design and virtual try-on applications.

NeRF-like methods learn implicit 3D neural representations from 2D multiview images, enabling the synthesis of compelling novel views. However, to capture high-fidelity geometry, prior methods often rely on large-scale networks. This dependency hampers the potential applications of neural implicit representations, such as MR visualization. To address this, we introduce LODNeuS, an implicit surface representation based on feature voxel grids. LODNeuS captures multiple LODs of implicit geometry by maintaining voxel grids paired with a set of corresponding lightweight decoders. This allows for high-quality rendering with the ability to dynamically switch between detail levels. Another challenge is that existing methods, both volumetric and surface-based, tend to train and render their representations within a confined space, without explicitly restricting the sampling points properly. This lack of constraints can result in ambiguity, artifacts, and inefficient use of computational resources. We study this effect during free viewpoint rendering using conventional methods and develop an adaptive sampling scheme that emphasizes a valid geometric space for sampling point allocation. Our experimental results show that LODNeuS can match the visual quality of existing methods while offering flexible and lightweight inference. The benefits of adaptive sampling are also demonstrated in the free viewpoint rendering subsection. Our work extends the capabilities of neural implicit representations beyond previously defined limitations, broadening the scope of potential applications.

Dataset distillation (DD) aims to synthesize a more compact dataset than the original one and models trained on it are expected to have the same generalization capabilities as on the original dataset. Previous work via a generative model (GM) faces several limitations. First, GM struggles to generate representative samples due to a lack of constraints. Second, it overlooks the relationships between generated samples, limiting its effectiveness. In this paper, a new noise-unconstrained GM-based DD framework is proposed. In the distillation stage, an adaptive matching coefficient is introduced to align generated images with representative class elements and the MiniMax loss function is extended to reduce the optimization difficulty. In the deployment stage, features among each generative image are ensembled by gradient-matching based DD. Theoretical analysis based on McDiarmid's inequality demonstrates that the proposed components can reduce the generalization error of the original baseline method. We also provide insights into the potential of generated images as an effective proxy dataset for DD. For example, on the ImageWoof dataset with 50 distilled images per class using a 6-layer ConvNet for evaluation, generated images outperform 25%, 50%, and 75% original images by 8.4%, 6.3%, and 8.3% in distillation performance. Our method effectively handles both low- and high-resolution datasets, with experiments on 11 benchmarks demonstrating its efficacy.

Recent advances in generative video models have blurred the boundary between real and synthetic content, raising urgent concerns about digital authenticity. Multimodal large language models (MLLMs) are appealing for AI-generated video (AIGV) forensics due to their broad perceptual and reasoning capabilities; however, existing MLLM-based detectors still suffer from hallucination and unstable reasoning, yielding high false-alarm rates and generic, non-verifiable explanations. To address these issues, we propose Hermes, an evidence-driven agentic framework for trustworthy and explainable AIGV detection. Hermes is realized by three key capabilities: (1) Adaptive Instance-Conditioned Detection Strategy Planning, (2) Evidence-Centric Reasoning and Verification, and (3) Graph-Grounded Evidence Deliberation. Concretely, Hermes employs an instance-conditioned RAG mechanism to analyze each video and retrieve authenticity-verification knowledge for composing a tailored detection strategy. It then performs evidence-centric reasoning by constructing a verifiable Evidence Reasoning Graph (ERG) that maintains focus on authenticity verification and avoids attention drift or superficial reasoning. Finally, a multi-agent deliberation process audits and refines the ERG to reconcile conflicting evidence and enhance reliability. Supported by these capabilities and a rich library of internal and external forensic tools, Hermes achieves structured, verifiable, and interpretable decision-making. Extensive experiments show that Hermes delivers state-of-the-art performance while producing higher-quality, auditable explanations.

High-precision scene parsing tasks, including image matting and dichotomous segmentation, aim to accurately predict masks with extremely fine details (such as hair). Most existing methods focus on salient, single foreground objects. While interactive methods allow for target adjustment, their class-agnostic design restricts generalization across different categories. Furthermore, the scarcity of high-quality annotation has led to a reliance on inharmonious synthetic data, resulting in poor generalization to real-world scenarios. To this end, we propose a Foreground Consistent Learning model, dubbed as FCLM, to address the aforementioned issues. Specifically, we first introduce a Depth-Aware Distillation strategy where we transfer the depth-related knowledge for better foreground representation. Considering the data dilemma, we term the processing of synthetic data as domain adaptation problem where we propose a domain-invariant learning strategy to focus on foreground learning. To support interactive prediction, we contribute an Object-Oriented Decoder that can receive both visual and language prompts to predict the referring target. Experimental results show that our method quantitatively and qualitatively outperforms state-of-the-art methods.

Traditional single-model approaches can achieve good performance in single-weather-affected images, but usually fail to restore real-world degraded images. This can be attributed to their reliance on weather-specific priors, while real-world weather degradation is typically caused by a combination of weather conditions. To address this limitation, we propose a Multi-Component Decomposition Network (MCD-Net), introducing a novel three-component prior designed to represent background content, primary degradation masks, and residual degradations separately. The proposed network employs multiple attention-guided branches to disentangle different degradation components from the multi-scale features extracted by the Transformer-based trunk network. Then the trunk output is decomposed into grouped features and dynamically assigned to branches with a fully connected layer to jointly estimate the clean background and the degraded input. Comprehensive experiments on four types of single-weather datasets and one multiple-weather degraded dataset validate that our MCD-Net significantly outperforms state-of-the-art methods, especially under challenging mixed-weather conditions.

Haze removal remains a challenging problem, especially in complex scenarios involving occlusion and varying fog intensities. Most existing methods perform well on specific datasets or scenes. However, they often fail to generalize because they focus only on haze regions, ignoring their interaction with surrounding objects. To address this issue, we propose a contrastive learning framework that works dynamically at the patch level. It captures the relationships between hazy and haze-free regions to handle varying fog intensities. The framework uses brightness differences between hazy and haze-free images to evaluate fog intensity. Based on this evaluation, the framework dynamically selects appropriate positive and negative samples, enhancing the model’s flexibility and efficacy. To further improve the robustness, we adopt a gray-scale image processing mechanism to convert RGB images to grayscale images. This reduces the light source interference and enables the network to focus on structural and contrast details. In addition, our novel feature extraction module combines long-range and short-range attention mechanisms to preserve the structural integrity of neighboring patches while modeling long-range dependencies. Experimental results show that our method can achieve better performance than that of the state-of-the-art approaches.

The manually designed neural networks have the drawbacks of requiring a large amount of training data and high computational costs. In this paper, we propose the masked autoencoder based lightweight network search algorithm which leverages the efficient channel search algorithm and specific distillation strategy to obtain the optimal architecture. During SuperNet training process, we design the cross-token distillation and cross-architecture strategy. Token distillation strategy is used to enforce the similar representation obtained from different masks in one image. Architecture distillation strategy is used to fully utilize representation from the sampled subnetwork and use the feature maps from one image but the same token. In the subnetwork searching process, we further pretrain the selected network considering the component dependency. Comprehensive experiments verify that our proposed method is efficient and flexible than baseline self-supervised learning algorithm and structured pruning algorithms. For example, our method obtains 4.4% improvement in TOP-1 metrics compared with the classic Masked Autoencoder algorithm designed lightweight Transformer architecture with less than 10M parameter.

Visual context is essential for point cloud semantic segmentation. The contextual information captures the semantic relationship between 3-D points, providing helpful hints for reasoning the category labels of points. Most current methods harness the internal context from the parts of the same object (or from the things within the same scene). In contrast, we propose the external point-set context (EPSC), allowing a richer context of external points learned across various objects/scenes to assist the segmentation task. We employ an external memory with multiple sets to store the EPSC representations learned from the training data. Each representation is a cluster feature, which captures the relationship between adjacent 3-D points for recognizing the semantic category of the center point. During the inference phase, the external memory releases the EPSC representations, providing rich and relevant context for segmenting the target point cloud. We extensively evaluate our method on Stanford Large-Scale 3-D Indoor Spaces (S3DIS), ScanNetv2, and ShapeNetPart datasets, where we achieve the result of effective improvement.

Images captured under rainy conditions often suffer from severe visual degradations, and pose significant challenges for high-level vision tasks. While deep learning-based approaches have shown promising performance, many of them rely on convolutional or attention-based architectures that either struggle to model global contextual information or suffer from high computational complexity. In this paper, we present FDRM-Net, a novel frequency-guided deraining framework based on Mamba. By leveraging the linear complexity and strong long-range modeling capabilities of state space models, our method efficiently captures spatial dependencies without the overhead of traditional self-attention mechanisms. In addition, we introduce frequency-aware representations to guide the learning process, enabling the network to better identify and remove structured rain patterns. A multi-scale architecture with progressive refinement further enhances detail recovery across varying rain densities. Extensive experiments demonstrate that FDRM-Net consistently achieves state-of-the-art performance in popular benchmarks, while offering favorable efficiency and robustness across diverse rain conditions.

The reuse of 3D CAD models is crucial for industrial manufacturing because it shortens development cycles and reduces costs. Significant progress has been made in deep learning-based 3D model retrievals. There are many representations for 3D models, among which the multi-view representation has demonstrated a superior retrieval performance. However, directly applying these 3D model retrieval approaches to 3D CAD model retrievals may result in issues such as the loss of the engineering semantic and structural information. In this paper, we find that multiple views and B-rep can complement each other. Therefore, we propose the view graph neural network (VGNet), which effectively combines multiple views and B-rep to accomplish 3D CAD model retrieval. More specifically, based on the characteristics of the regular shape of 3D CAD models, and the richness of the attribute information in the B-rep attribute graph, we separately design two feature extraction networks for each modality. Moreover, to explore the latent relationships between the multiple views and B-rep attribute graphs, a multi-head attention enhancement module is designed. Furthermore, the multimodal fusion module is adopted to make the joint representation of the 3D CAD models more discriminative by using a correlation loss function. Experiments are carried out on a real manufacturing 3D CAD dataset and a public dataset to validate the effectiveness of the proposed approach.

Nebulized therapy is essential for treating respiratory illnesses in children, yet it can elicit fear and resistance. To tackle this challenge, we developed FunBreath—a novel interactive nebulizer mask with a gamification system. By integrating both hardware and software components, we revolutionized the therapy experience. Our approach began with in-depth surveys to understand the needs of children and their parents in relation to nebulized therapy. Subsequently, for the software component, we adopted the mechanics-dynamics-aesthetics framework to create engaging interactive games. For the hardware component, we incorporated a pressure sensor to monitor children’s breathing behavior, dynamically influencing gameplay. User studies suggest that the FunBreath system could provide a more effective and enjoyable therapy experience, potentially improving treatment adherence by enhancing children’s motivation, increasing their participation, and reducing anxiety during treatment. This work enabled us to distill and contextualize pivotal gamification design principles for medical health products, paving the way for future advancements in this domain.

The human skeleton establishment aims to provide accurate localization information of the human body from RGB images and establish a complete human skeleton for many applications, such as action recognition, video surveillance, and human-computer interaction. Considering the inherent human body structure, many recent methods group the relevant body parts and utilize the deep convolutional network to learn the visual context from the part groups. However, the grouping approaches used in these methods heavily rely on prior knowledge of the human body shape but lose important relationships between parts. In this paper, we introduce the Accurate Part Grouping Network (Accurate-PGNet), a novel network for hierarchically grouping body parts in a data-driven manner. In contrast to the previous methods, we use neural architecture search (NAS) to optimize the architecture of Accurate-PGNet and properly group the body parts. The part grouping respects the diverse visual patterns of parts, producing groups containing different body parts. From each group, we learn the visual feature map. It helps to capture the correlation between parts and predict their locations. The feature maps of the part groups are merged hierarchically to capture the higher-order context of parts in larger groups. We extensively evaluated our method on the challenging benchmarks, demonstrating that Accurate-PGNet effectively helps to achieve state-of-the-art results.

3D inverse rendering in indoor scenes with strong light sources presents a significant challenge, primarily due to the substantial ambiguity in material recovery caused by the complex interaction between lighting and shadows. To address this, we propose a novel approach that integrates an implicit-explicit shadow predictor with a three-stage material estimation process. Our method enhances shadow realism by accurately predicting light interactions, while our material estimation process improves SVBRDF quality under challenging lighting conditions. Extensive experiments demonstrate the effectiveness of our method in both quantitative and qualitative metrics, enabling realistic object insertion and material replacement with proper shadow rendering under strong indoor light sources.

Convolutional neural networks (CNNs) are widely used for embroidery feature synthesis from images. However, they are still unable to predict diverse stitch types, which makes it difficult for the CNNs to effectively extract stitch features. In this paper, we propose a multi-stitch embroidery generative adversarial network (MSEmbGAN) that uses a region-aware texture generation sub-network to predict diverse embroidery features from images. To the best of our knowledge, our work is the first CNN-based generative adversarial network to succeed in this task. Our region-aware texture generation sub-network detects multiple regions in the input image using a stitch classifier and generates a stitch texture for each region based on its shape features. We also propose a colorization network with a color feature extractor, which helps achieve full image color consistency by requiring the color attributes of the output to closely resemble the input image. Because of the current lack of labeled embroidery image datasets, we provide a new multi-stitch embroidery dataset that is annotated with three single-stitch types and one multi-stitch type. Our dataset, which includes more than 30K high-quality multi-stitch embroidery images, more than 13K aligned content-embroidered images, and more than 17K unaligned images, is currently the largest embroidery dataset accessible, as far as we know. Quantitative and qualitative experimental results, including a qualitative user study, show that our MSEmbGAN outperforms current state-of-the-art embroidery synthesis and style-transfer methods on all evaluation indicators.

Multi-domain image inpainting utilizes complementary contextual information from auxiliary domain images to restore corrupted regions. While existing methods reconstruct auxiliary images to provide additional guidance, they face fundamental limitations: recovered pixels with complex patterns often lack representative details, while oversimplified patterns offer insufficient contextual information. To address these challenges, we propose HRC-Net, a novel framework incorporating three generative sub-networks for the comprehensive image inpainting task. Our architecture consists of: (1) A Hypothesis Sub-network that enables robust samplings of pixel-wise hypotheses from multi-domain inputs; (2) A Representative Sub-network that learns to score hypothesis quality based on contextual relevance; and (3) a Collaboration Sub-network that optimizes adaptive fusion kernels to integrate the most pertinent details. Together, these components model the joint distribution of representative scores and convolutional kernels, fostering a precise interaction between auxiliary hypotheses and target image corruption to meticulously repair the target image. Extensive evaluations across multiple benchmark datasets demonstrate HRC-Net's superior performance, significantly outperforming state-of-the-art methods in both quantitative metrics and visual quality.