Zhaoxin Fan

Xueting Yang, Yihao Luo, Yuliang Xiu et al.

Realistic virtual humans play a crucial role in numerous industries, such as metaverse, intelligent healthcare, and self-driving simulation. But creating them on a large scale with high levels of realism remains a challenge. The utilization of deep implicit function sparks a new era of image-based 3D clothed human reconstruction, enabling pixel-aligned shape recovery with fine details. Subsequently, the vast majority of works locate the surface by regressing the deterministic implicit value for each point. However, should all points be treated equally regardless of their proximity to the surface? In this paper, we propose replacing the implicit value with an adaptive uncertainty distribution, to differentiate between points based on their distance to the surface. This simple ``value to distribution'' transition yields significant improvements on nearly all the baselines. Furthermore, qualitative results demonstrate that the models trained using our uncertainty distribution loss, can capture more intricate wrinkles, and realistic limbs. Code and models are available for research purposes at https://github.com/psyai-net/D-IF_release.

Zhaoxin Fan, Kaixing Yang, Min Zhang et al.

Recently, over-height vehicle strike frequently occurs, causing great economic cost and serious safety problems. Hence, an alert system which can accurately discover any possible height limiting devices in advance is necessary to be employed in modern large or medium sized cars, such as touring cars. Detecting and estimating the height limiting devices act as the key point of a successful height limit alert system. Though there are some works research height limit estimation, existing methods are either too computational expensive or not accurate enough. In this paper, we propose a novel stereo-based pipeline named SHLE for height limit estimation. Our SHLE pipeline consists of two stages. In stage 1, a novel devices detection and tracking scheme is introduced, which accurately locate the height limit devices in the left or right image. Then, in stage 2, the depth is temporally measured, extracted and filtered to calculate the height limit device. To benchmark the height limit estimation task, we build a large-scale dataset named "Disparity Height", where stereo images, pre-computed disparities and ground-truth height limit annotations are provided. We conducted extensive experiments on "Disparity Height" and the results show that SHLE achieves an average error below than 10cm though the car is 70m away from the devices. Our method also outperforms all compared baselines and achieves state-of-the-art performance. Code is available at https://github.com/Yang-Kaixing/SHLE.

Zhenyuan Zhang, Zhenbo Song, Kaihao Zhang et al.

Deep learning based methods have achieved significant success in the task of single image reflection removal (SIRR). However, the majority of these methods are focused on High-Definition/Standard-Definition (HD/SD) images, while ignoring higher resolution images such as Ultra-High-Definition (UHD) images. With the increasing prevalence of UHD images captured by modern devices, in this paper, we aim to address the problem of UHD SIRR. Specifically, we first synthesize two large-scale UHD datasets, UHDRR4K and UHDRR8K. The UHDRR4K dataset consists of $2,999$ and $168$ quadruplets of images for training and testing respectively, and the UHDRR8K dataset contains $1,014$ and $105$ quadruplets. To the best of our knowledge, these two datasets are the first largest-scale UHD datasets for SIRR. Then, we conduct a comprehensive evaluation of six state-of-the-art SIRR methods using the proposed datasets. Based on the results, we provide detailed discussions regarding the strengths and limitations of these methods when applied to UHD images. Finally, we present a transformer-based architecture named RRFormer for reflection removal. RRFormer comprises three modules, namely the Prepossessing Embedding Module, Self-attention Feature Extraction Module, and Multi-scale Spatial Feature Extraction Module. These modules extract hypercolumn features, global and partial attention features, and multi-scale spatial features, respectively. To ensure effective training, we utilize three terms in our loss function: pixel loss, feature loss, and adversarial loss. We demonstrate through experimental results that RRFormer achieves state-of-the-art performance on both the non-UHD dataset and our proposed UHDRR datasets. The code and datasets are publicly available at https://github.com/Liar-zzy/Benchmarking-Ultra-High-Definition-Single-Image-Reflection-Removal.

Xin Li, Yeying Jin, Suhang Yao et al.

This paper presents an overview of the NTIRE 2026 Second Challenge on Day and Night Raindrop Removal for Dual-Focused Images. Building upon the success of the first edition, this challenge attracted a wide range of impressive solutions, all developed and evaluated on our real-world Raindrop Clarity dataset~\cite{jin2024raindrop}. For this edition, we adjust the dataset with 14,139 images for training, 407 images for validation, and 593 images for testing. The primary goal of this challenge is to establish a strong and practical benchmark for the removal of raindrops under various illumination and focus conditions. In total, 168 teams have registered for the competition, and 17 teams submitted valid final solutions and fact sheets for the testing phase. The submitted methods achieved strong performance on the Raindrop Clarity dataset, demonstrating the growing progress in this challenging task.

Han Sun, Zhaoxin Fan, Zhenbo Song et al.

Monocular 3D object detection is a fundamental but very important task to many applications including autonomous driving, robotic grasping and augmented reality. Existing leading methods tend to estimate the depth of the input image first, and detect the 3D object based on point cloud. This routine suffers from the inherent gap between depth estimation and object detection. Besides, the prediction error accumulation would also affect the performance. In this paper, a novel method named MonoSIM is proposed. The insight behind introducing MonoSIM is that we propose to simulate the feature learning behaviors of a point cloud based detector for monocular detector during the training period. Hence, during inference period, the learned features and prediction would be similar to the point cloud based detector as possible. To achieve it, we propose one scene-level simulation module, one RoI-level simulation module and one response-level simulation module, which are progressively used for the detector's full feature learning and prediction pipeline. We apply our method to the famous M3D-RPN detector and CaDDN detector, conducting extensive experiments on KITTI and Waymo Open datasets. Results show that our method consistently improves the performance of different monocular detectors for a large margin without changing their network architectures. Our codes will be publicly available at https://github.com/sunh18/MonoSIM}{https://github.com/sunh18/MonoSIM.

Ziqiao Peng, Haoyu Wu, Zhenbo Song et al.

Speech-driven 3D face animation aims to generate realistic facial expressions that match the speech content and emotion. However, existing methods often neglect emotional facial expressions or fail to disentangle them from speech content. To address this issue, this paper proposes an end-to-end neural network to disentangle different emotions in speech so as to generate rich 3D facial expressions. Specifically, we introduce the emotion disentangling encoder (EDE) to disentangle the emotion and content in the speech by cross-reconstructed speech signals with different emotion labels. Then an emotion-guided feature fusion decoder is employed to generate a 3D talking face with enhanced emotion. The decoder is driven by the disentangled identity, emotional, and content embeddings so as to generate controllable personal and emotional styles. Finally, considering the scarcity of the 3D emotional talking face data, we resort to the supervision of facial blendshapes, which enables the reconstruction of plausible 3D faces from 2D emotional data, and contribute a large-scale 3D emotional talking face dataset (3D-ETF) to train the network. Our experiments and user studies demonstrate that our approach outperforms state-of-the-art methods and exhibits more diverse facial movements. We recommend watching the supplementary video: https://ziqiaopeng.github.io/emotalk

Ziqiao Peng, Yihao Luo, Yue Shi et al.

Speech-driven 3D face animation technique, extending its applications to various multimedia fields. Previous research has generated promising realistic lip movements and facial expressions from audio signals. However, traditional regression models solely driven by data face several essential problems, such as difficulties in accessing precise labels and domain gaps between different modalities, leading to unsatisfactory results lacking precision and coherence. To enhance the visual accuracy of generated lip movement while reducing the dependence on labeled data, we propose a novel framework SelfTalk, by involving self-supervision in a cross-modals network system to learn 3D talking faces. The framework constructs a network system consisting of three modules: facial animator, speech recognizer, and lip-reading interpreter. The core of SelfTalk is a commutative training diagram that facilitates compatible features exchange among audio, text, and lip shape, enabling our models to learn the intricate connection between these factors. The proposed framework leverages the knowledge learned from the lip-reading interpreter to generate more plausible lip shapes. Extensive experiments and user studies demonstrate that our proposed approach achieves state-of-the-art performance both qualitatively and quantitatively. We recommend watching the supplementary video.

Ziqiao Peng, Wentao Hu, Yue Shi et al.

Achieving high synchronization in the synthesis of realistic, speech-driven talking head videos presents a significant challenge. Traditional Generative Adversarial Networks (GAN) struggle to maintain consistent facial identity, while Neural Radiance Fields (NeRF) methods, although they can address this issue, often produce mismatched lip movements, inadequate facial expressions, and unstable head poses. A lifelike talking head requires synchronized coordination of subject identity, lip movements, facial expressions, and head poses. The absence of these synchronizations is a fundamental flaw, leading to unrealistic and artificial outcomes. To address the critical issue of synchronization, identified as the "devil" in creating realistic talking heads, we introduce SyncTalk. This NeRF-based method effectively maintains subject identity, enhancing synchronization and realism in talking head synthesis. SyncTalk employs a Face-Sync Controller to align lip movements with speech and innovatively uses a 3D facial blendshape model to capture accurate facial expressions. Our Head-Sync Stabilizer optimizes head poses, achieving more natural head movements. The Portrait-Sync Generator restores hair details and blends the generated head with the torso for a seamless visual experience. Extensive experiments and user studies demonstrate that SyncTalk outperforms state-of-the-art methods in synchronization and realism. We recommend watching the supplementary video: https://ziqiaopeng.github.io/synctalk

Zhaoxin Fan, Zhenbo Song, Jian Xu et al.

Recently, RGBD-based category-level 6D object pose estimation has achieved promising improvement in performance, however, the requirement of depth information prohibits broader applications. In order to relieve this problem, this paper proposes a novel approach named Object Level Depth reconstruction Network (OLD-Net) taking only RGB images as input for category-level 6D object pose estimation. We propose to directly predict object-level depth from a monocular RGB image by deforming the category-level shape prior into object-level depth and the canonical NOCS representation. Two novel modules named Normalized Global Position Hints (NGPH) and Shape-aware Decoupled Depth Reconstruction (SDDR) module are introduced to learn high fidelity object-level depth and delicate shape representations. At last, the 6D object pose is solved by aligning the predicted canonical representation with the back-projected object-level depth. Extensive experiments on the challenging CAMERA25 and REAL275 datasets indicate that our model, though simple, achieves state-of-the-art performance.

Zhaoxin Fan, Yulin He, Zhicheng Wang et al.

Real-world sensors often produce incomplete, irregular, and noisy point clouds, making point cloud completion increasingly important. However, most existing completion methods rely on large paired datasets for training, which is labor-intensive. This paper proposes RaPD, a novel semi-supervised point cloud completion method that reduces the need for paired datasets. RaPD utilizes a two-stage training scheme, where a deep semantic prior is learned in stage 1 from unpaired complete and incomplete point clouds, and a semi-supervised prior distillation process is introduced in stage 2 to train a completion network using only a small number of paired samples. Additionally, a self-supervised completion module is introduced to improve performance using unpaired incomplete point clouds. Experiments on multiple datasets show that RaPD outperforms previous methods in both homologous and heterologous scenarios.

Xukun Zhou, Fengxin Li, Ziqiao Peng et al.

Audio-driven 3D face animation is increasingly vital in live streaming and augmented reality applications. While remarkable progress has been observed, most existing approaches are designed for specific individuals with predefined speaking styles, thus neglecting the adaptability to varied speaking styles. To address this limitation, this paper introduces MetaFace, a novel methodology meticulously crafted for speaking style adaptation. Grounded in the novel concept of meta-learning, MetaFace is composed of several key components: the Robust Meta Initialization Stage (RMIS) for fundamental speaking style adaptation, the Dynamic Relation Mining Neural Process (DRMN) for forging connections between observed and unobserved speaking styles, and the Low-rank Matrix Memory Reduction Approach to enhance the efficiency of model optimization as well as learning style details. Leveraging these novel designs, MetaFace not only significantly outperforms robust existing baselines but also establishes a new state-of-the-art, as substantiated by our experimental results.

Haoyu Wu, Ziqiao Peng, Xukun Zhou et al.

3D face reconstruction from monocular images has promoted the development of various applications such as augmented reality. Though existing methods have made remarkable progress, most of them emphasize geometric reconstruction, while overlooking the importance of texture prediction. To address this issue, we propose VGG-Tex, a novel Vivid Geometry-Guided Facial Texture Estimation model designed for High Fidelity Monocular 3D Face Reconstruction. The core of this approach is leveraging 3D parametric priors to enhance the outcomes of 2D UV texture estimation. Specifically, VGG-Tex includes a Facial Attributes Encoding Module, a Geometry-Guided Texture Generator, and a Visibility-Enhanced Texture Completion Module. These components are responsible for extracting parametric priors, generating initial textures, and refining texture details, respectively. Based on the geometry-texture complementarity principle, VGG-Tex also introduces a Texture-guided Geometry Refinement Module to further balance the overall fidelity of the reconstructed 3D faces, along with corresponding losses. Comprehensive experiments demonstrate that our method significantly improves texture reconstruction performance compared to existing state-of-the-art methods.

Zhaoxin Fan, Zhenbo Song, Hongyan Liu et al.

Large-scale place recognition is a fundamental but challenging task, which plays an increasingly important role in autonomous driving and robotics. Existing methods have achieved acceptable good performance, however, most of them are concentrating on designing elaborate global descriptor learning network structures. The importance of feature generalization and descriptor post-enhancing has long been neglected. In this work, we propose a novel method named GIDP to learn a Good Initialization and Inducing Descriptor Poseenhancing for Large-scale Place Recognition. In particular, an unsupervised momentum contrast point cloud pretraining module and a reranking-based descriptor post-enhancing module are proposed respectively in GIDP. The former aims at learning a good initialization for the point cloud encoding network before training the place recognition model, while the later aims at post-enhancing the predicted global descriptor through reranking at inference time. Extensive experiments on both indoor and outdoor datasets demonstrate that our method can achieve state-of-the-art performance using simple and general point cloud encoding backbones.

Zhaoxin Fan, Fengxin Li, Hongyan Liu et al.

In this paper, we research the new topic of object effects recommendation in micro-video platforms, which is a challenging but important task for many practical applications such as advertisement insertion. To avoid the problem of introducing background bias caused by directly learning video content from image frames, we propose to utilize the meaningful body language hidden in 3D human pose for recommendation. To this end, in this work, a novel human pose driven object effects recommendation network termed PoseRec is introduced. PoseRec leverages the advantages of 3D human pose detection and learns information from multi-frame 3D human pose for video-item registration, resulting in high quality object effects recommendation performance. Moreover, to solve the inherent ambiguity and sparsity issues that exist in object effects recommendation, we further propose a novel item-aware implicit prototype learning module and a novel pose-aware transductive hard-negative mining module to better learn pose-item relationships. What's more, to benchmark methods for the new research topic, we build a new dataset for object effects recommendation named Pose-OBE. Extensive experiments on Pose-OBE demonstrate that our method can achieve superior performance than strong baselines.

Zhaoxin Fan, Yuqing Pan, Hao Xu et al.

In the field of full-body reconstruction, the scarcity of annotated data often impedes the efficacy of prevailing methods. To address this issue, we introduce FuRPE, a novel framework that employs part-experts and an ingenious pseudo ground-truth selection scheme to derive high-quality pseudo labels. These labels, central to our approach, equip our network with the capability to efficiently learn from the available data. Integral to FuRPE is a unique exponential moving average training strategy and expert-derived feature distillation strategy. These novel elements of FuRPE not only serve to further refine the model but also to reduce potential biases that may arise from inaccuracies in pseudo labels, thereby optimizing the network's training process and enhancing the robustness of the model. We apply FuRPE to train both two-stage and fully convolutional single-stage full-body reconstruction networks. Our exhaustive experiments on numerous benchmark datasets illustrate a substantial performance boost over existing methods, underscoring FuRPE's potential to reshape the state-of-the-art in full-body reconstruction.

Zirui Ge, Pengxiang Ding, Baohua Yin et al.

Video action models are an appealing foundation for Vision--Language--Action systems because they can learn visual dynamics from large-scale video data and transfer this knowledge to downstream robot control. Yet current diffusion-based video predictors are trained with likelihood-surrogate objectives, which encourage globally plausible predictions without explicitly optimizing the precision-critical visual dynamics needed for manipulation. This objective mismatch often leads to subtle errors in object pose, spatial relations, and contact timing that can be amplified by downstream policies. We propose VAMPO, a post-training framework that directly improves visual dynamics in video action models through policy optimization. Our key idea is to formulate multi-step denoising as a sequential decision process and optimize the denoising policy with rewards defined over expert visual dynamics in latent space. To make this optimization practical, we introduce an Euler Hybrid sampler that injects stochasticity only at the first denoising step, enabling tractable low-variance policy-gradient estimation while preserving the coherence of the remaining denoising trajectory. We further combine this design with GRPO and a verifiable non-adversarial reward. Across diverse simulated and real-world manipulation tasks, VAMPO improves task-relevant visual dynamics, leading to better downstream action generation and stronger generalization. The homepage is https://vampo-robot.github.io/VAMPO/.

Xukun Zhou, Zhenbo Song, Jun He et al.

Scene Graph Generation is a critical enabler of environmental comprehension for autonomous robotic systems. Most of existing methods, however, are often thwarted by the intricate dynamics of background complexity, which limits their ability to fully decode the inherent topological information of the environment. Additionally, the wealth of contextual information encapsulated within depth cues is often left untapped, rendering existing approaches less effective. To address these shortcomings, we present STDG, an avant-garde Depth-Guided One-Stage Scene Graph Generation methodology. The innovative architecture of STDG is a triad of custom-built modules: The Depth Guided HHA Representation Generation Module, the Depth Guided Semi-Teaching Network Learning Module, and the Depth Guided Scene Graph Generation Module. This trifecta of modules synergistically harnesses depth information, covering all aspects from depth signal generation and depth feature utilization, to the final scene graph prediction. Importantly, this is achieved without imposing additional computational burden during the inference phase. Experimental results confirm that our method significantly enhances the performance of one-stage scene graph generation baselines.

Nanxiang Jiang, Zhaoxin Fan, Baisen Wang et al.

Concept erasure serves as a vital safety mechanism for removing unwanted concepts from text-to-image (T2I) models. While extensively studied in U-Net and dual-stream architectures (e.g., Flux), this task remains under-explored in the recent emerging paradigm of single-stream diffusion transformers (e.g., Z-Image). In this new paradigm, text and image tokens are processed as a single unified sequence via shared parameters. Consequently, directly applying prior erasure methods typically leads to generation collapse. To bridge this gap, we introduce Z-Erase, the first concept erasure method tailored for single-stream T2I models. To guarantee stable image generation, Z-Erase first proposes a Stream Disentangled Concept Erasure Framework that decouples updates and enables existing methods on single-stream models. Subsequently, within this framework, we introduce Lagrangian-Guided Adaptive Erasure Modulation, a constrained algorithm that further balances the sensitive erasure-preservation trade-off. Moreover, we provide a rigorous convergence analysis proving that Z-Erase can converge to a Pareto stationary point. Experiments demonstrate that Z-Erase successfully overcomes the generation collapse issue, achieving state-of-the-art performance across a wide range of tasks.

Hongzhi Zhang, Yuanze Hu, Tinghai Zhang et al.

The evolution of Large Language Models (LLMs) towards autonomous agents has catalyzed progress in Deep Research. While retrieval capabilities are well-benchmarked, the post-retrieval synthesis stage--where agents must digest massive amounts of context and consolidate fragmented evidence into coherent, long-form reports--remains under-evaluated due to the subjectivity of open-ended writing. To bridge this gap, we introduce DeepSynth-Eval, a benchmark designed to objectively evaluate information consolidation capabilities. We leverage high-quality survey papers as gold standards, reverse-engineering research requests and constructing "Oracle Contexts" from their bibliographies to isolate synthesis from retrieval noise. We propose a fine-grained evaluation protocol using General Checklists (for factual coverage) and Constraint Checklists (for structural organization), transforming subjective judgment into verifiable metrics. Experiments across 96 tasks reveal that synthesizing information from hundreds of references remains a significant challenge. Our results demonstrate that agentic plan-and-write workflows significantly outperform single-turn generation, effectively reducing hallucinations and improving adherence to complex structural constraints.

Shuo Wang, Yucheng Wang, Guoxin Lian et al.

Vision-Language Navigation requires agents to act coherently over long horizons by understanding not only local visual context but also how far they have advanced within a multi-step instruction. However, recent Vision-Language-Action models focus on direct action prediction and earlier progress methods predict numeric achievements; both overlook the monotonic co-progression property of the observation and instruction sequences. Building on this insight, Progress-Think introduces semantic progress reasoning, predicting instruction-style progress from visual observations to enable more accurate navigation. To achieve this without expensive annotations, we propose a three-stage framework. In the initial stage, Self-Aligned Progress Pretraining bootstraps a reasoning module via a novel differentiable alignment between visual history and instruction prefixes. Then, Progress-Guided Policy Pretraining injects learned progress states into the navigation context, guiding the policy toward consistent actions. Finally, Progress-Policy Co-Finetuning jointly optimizes both modules with tailored progress-aware reinforcement objectives. Experiments on R2R-CE and RxR-CE show state-of-the-art success and efficiency, demonstrating that semantic progress yields a more consistent representation of navigation advancement.

Yixing Lu, Zhaoxin Fan, Min Xu

In this paper, we introduce a novel semi-supervised learning framework tailored for medical image segmentation. Central to our approach is the innovative Multi-scale Text-aware ViT-CNN Fusion scheme. This scheme adeptly combines the strengths of both ViTs and CNNs, capitalizing on the unique advantages of both architectures as well as the complementary information in vision-language modalities. Further enriching our framework, we propose the Multi-Axis Consistency framework for generating robust pseudo labels, thereby enhancing the semisupervised learning process. Our extensive experiments on several widelyused datasets unequivocally demonstrate the efficacy of our approach.

Zhaoxin Fan, Puquan Pan, Zeren Zhang et al.

Few-shot medical image semantic segmentation is of paramount importance in the domain of medical image analysis. However, existing methodologies grapple with the challenge of data scarcity during the training phase, leading to over-fitting. To mitigate this issue, we introduce a novel Unsupervised Dense Few-shot Medical Image Segmentation Model Training Pipeline (DenseMP) that capitalizes on unsupervised dense pre-training. DenseMP is composed of two distinct stages: (1) segmentation-aware dense contrastive pre-training, and (2) few-shot-aware superpixel guided dense pre-training. These stages collaboratively yield a pre-trained initial model specifically designed for few-shot medical image segmentation, which can subsequently be fine-tuned on the target dataset. Our proposed pipeline significantly enhances the performance of the widely recognized few-shot segmentation model, PA-Net, achieving state-of-the-art results on the Abd-CT and Abd-MRI datasets. Code will be released after acceptance.

Mingjin Chen, Junhao Chen, Zhaoxin Fan et al.

Recent methods have made notable progress in the visual quality of hand-object interaction video synthesis. However, most approaches rely on 2D control signals that lack spatial expressiveness and limit the utilization of synthetic 3D conditional data. To address these limitations, we propose HVG-3D, a unified framework for 3D-aware hand-object interaction (HOI) video synthesis conditioned on explicit 3D representations. Specifically, we develop a diffusion-based architecture augmented with a 3D ControlNet, which encodes geometric and motion cues from 3D inputs to enable explicit 3D reasoning during video synthesis. To achieve high-quality synthesis, HVG-3D is designed with two core components: (i) a 3D-aware HOI video generation diffusion architecture that encodes geometric and motion cues from 3D inputs for explicit 3D reasoning; and (ii) a hybrid pipeline for constructing input and condition signals, enabling flexible and precise control during both training and inference. During inference, given a single real image and a 3D control signal from either simulation or real data, HVG-3D generates high-fidelity, temporally consistent videos with precise spatial and temporal control. Experiments on the TASTE-Rob dataset demonstrate that HVG-3D achieves state-of-the-art spatial fidelity, temporal coherence, and controllability, while enabling effective utilization of both real and simulated data.

Ziqiao Peng, Yi Chen, Yifeng Ma et al.

Despite significant advances in talking avatar generation, existing methods face critical challenges: insufficient text-following capability for diverse actions, lack of temporal alignment between actions and audio content, and dependency on additional control signals such as pose skeletons. We present ActAvatar, a framework that achieves phase-level precision in action control through textual guidance by capturing both action semantics and temporal context. Our approach introduces three core innovations: (1) Phase-Aware Cross-Attention (PACA), which decomposes prompts into a global base block and temporally-anchored phase blocks, enabling the model to concentrate on phase-relevant tokens for precise temporal-semantic alignment; (2) Progressive Audio-Visual Alignment, which aligns modality influence with the hierarchical feature learning process-early layers prioritize text for establishing action structure while deeper layers emphasize audio for refining lip movements, preventing modality interference; (3) A two-stage training strategy that first establishes robust audio-visual correspondence on diverse data, then injects action control through fine-tuning on structured annotations, maintaining both audio-visual alignment and the model's text-following capabilities. Extensive experiments demonstrate that ActAvatar significantly outperforms state-of-the-art methods in both action control and visual quality.

Yihong Lin, Zhaoxin Fan, Xianjia Wu et al.

Speech-driven talking head generation is a critical yet challenging task with applications in augmented reality and virtual human modeling. While recent approaches using autoregressive and diffusion-based models have achieved notable progress, they often suffer from modality inconsistencies, particularly misalignment between audio and mesh, leading to reduced motion diversity and lip-sync accuracy. To address this, we propose GLDiTalker, a novel speech-driven 3D facial animation model based on a Graph Latent Diffusion Transformer. GLDiTalker resolves modality misalignment by diffusing signals within a quantized spatiotemporal latent space. It employs a two-stage training pipeline: the Graph-Enhanced Quantized Space Learning Stage ensures lip-sync accuracy, while the Space-Time Powered Latent Diffusion Stage enhances motion diversity. Together, these stages enable GLDiTalker to generate realistic, temporally stable 3D facial animations. Extensive evaluations on standard benchmarks demonstrate that GLDiTalker outperforms existing methods, achieving superior results in both lip-sync accuracy and motion diversity.

Runmin Jiang, Zhaoxin Fan, Junhao Wu et al.

3D medical image segmentation is a challenging task with crucial implications for disease diagnosis and treatment planning. Recent advances in deep learning have significantly enhanced fully supervised medical image segmentation. However, this approach heavily relies on labor-intensive and time-consuming fully annotated ground-truth labels, particularly for 3D volumes. To overcome this limitation, we propose a novel probabilistic-aware weakly supervised learning pipeline, specifically designed for 3D medical imaging. Our pipeline integrates three innovative components: a Probability-based Pseudo Label Generation technique for synthesizing dense segmentation masks from sparse annotations, a Probabilistic Multi-head Self-Attention network for robust feature extraction within our Probabilistic Transformer Network, and a Probability-informed Segmentation Loss Function to enhance training with annotation confidence. Demonstrating significant advances, our approach not only rivals the performance of fully supervised methods but also surpasses existing weakly supervised methods in CT and MRI datasets, achieving up to 18.1% improvement in Dice scores for certain organs. The code is available at https://github.com/runminjiang/PW4MedSeg.

Guoxin Lian, Shuo Wang, Yucheng Wang et al.

Vision-Language Navigation (VLN) requires agents to follow natural language instructions in partially observed 3D environments, motivating map representations that aggregate spatial context beyond local perception. However, most existing approaches rely on hand-crafted maps constructed independently of the navigation policy. We argue that maps should instead be learned representations shaped directly by navigation objectives rather than exhaustive reconstructions. Based on this insight, we propose MapDream, a map-in-the-loop framework that formulates map construction as autoregressive bird's-eye-view (BEV) image synthesis. The framework jointly learns map generation and action prediction, distilling environmental context into a compact three-channel BEV map that preserves only navigation-critical affordances. Supervised pre-training bootstraps a reliable mapping-to-control interface, while the autoregressive design enables end-to-end joint optimization through reinforcement fine-tuning. Experiments on R2R-CE and RxR-CE achieve state-of-the-art monocular performance, validating task-driven generative map learning.

Yuqin Lan, Gen Li, Yuanze Hu et al.

Vision-Language Models (VLMs) are powerful but remain vulnerable to multimodal jailbreak attacks. Existing attacks mainly rely on either explicit visual prompt attacks or gradient-based adversarial optimization. While the former is easier to detect, the latter produces subtle perturbations that are less perceptible, but is usually optimized and evaluated under homogeneous open-source surrogate-target settings, leaving its effectiveness on commercial closed-source VLMs under heterogeneous settings unclear. To examine this issue, we study different surrogate-target settings and observe a consistent gap between homogeneous and heterogeneous settings, a phenomenon we term surrogate dependency. Motivated by this finding, we propose Mosaic, a Multi-view ensemble optimization framework for multimodal jailbreak against closed-source VLMs, which alleviates surrogate dependency under heterogeneous surrogate-target settings by reducing over-reliance on any single surrogate model and visual view. Specifically, Mosaic incorporates three core components: a Text-Side Transformation module, which perturbs refusal-sensitive lexical patterns; a Multi-View Image Optimization module, which updates perturbations under diverse cropped views to avoid overfitting to a single visual view; and a Surrogate Ensemble Guidance module, which aggregates optimization signals from multiple surrogate VLMs to reduce surrogate-specific bias. Extensive experiments on safety benchmarks demonstrate that Mosaic achieves state-of-the-art Attack Success Rate and Average Toxicity against commercial closed-source VLMs.

Tianbao Zhang, Zhenyu Liang, Zhenbo Song et al.

High-resolution geometric prediction is essential for robust perception in autonomous driving, robotics, and AR/MR, but current foundation models are fundamentally limited by their scalability to real-world, high-resolution scenarios. Direct inference on 2K images with these models incurs prohibitive computational and memory demands, making practical deployment challenging. To tackle the issue, we present 2K Retrofit, a novel framework that enables efficient 2K-resolution inference for any geometric foundation model, without modifying or retraining the backbone. Our approach leverages fast coarse predictions and an entropy-based sparse refinement to selectively enhance high-uncertainty regions, achieving precise and high-fidelity 2K outputs with minimal overhead. Extensive experiments on widely used benchmark demonstrate that 2K Retrofit consistently achieves state-of-the-art accuracy and speed, bridging the gap between research advances and scalable deployment in high-resolution 3D vision applications. Code will be released upon acceptance.

Boshui Chen, Zhaoxin Fan, Ke Wang et al.

Large Language Models (LLMs) are powerful and widely adopted, but their practical impact is limited by the well-known hallucination phenomenon. While recent hallucination detection methods have made notable progress, we find most of them overlook the dynamic nature and underlying mechanisms of it. To address this gap, we propose HalluSAE, a phase transition-inspired framework that models hallucination as a critical shift in the model's latent dynamics. By modeling the generation process as a trajectory through a potential energy landscape, HalluSAE identifies critical transition zones and attributes factual errors to specific high-energy sparse features. Our approach consists of three stages: (1) Potential Energy Empowered Phase Zone Localization via sparse autoencoders and a geometric potential energy metric; (2) Hallucination-related Sparse Feature Attribution using contrastive logit attribution; and (3) Probing-based Causal Hallucination Detection through linear probes on disentangled features. Extensive experiments on Gemma-2-9B demonstrate that HalluSAE achieves state-of-the-art hallucination detection performance.

Xiaomin Yu, Pengxiang Ding, Wenjie Zhang et al.

Training vision-language models (VLMs) typically requires large-scale, high-quality image-text pairs, but collecting or synthesizing such data is costly. In contrast, text data is abundant and inexpensive, prompting the question: can high-quality multimodal training data be synthesized purely from text? To tackle this, we propose a cross-integrated three-stage multimodal data synthesis framework, which generates two datasets: Unicorn-1.2M and Unicorn-471K-Instruction. In Stage 1: Diverse Caption Data Synthesis, we construct 1.2M semantically diverse high-quality captions by expanding sparse caption seeds using large language models (LLMs). In Stage 2: Instruction-Tuning Data Generation, we further process 471K captions into multi-turn instruction-tuning tasks to support complex reasoning. Finally, in Stage 3: Modality Representation Transfer, these textual captions representations are transformed into visual representations, resulting in diverse synthetic image representations. This three-stage process enables us to construct Unicorn-1.2M for pretraining and Unicorn-471K-Instruction for instruction-tuning, without relying on real images. By eliminating the dependency on real images while maintaining data quality and diversity, our framework offers a cost-effective and scalable solution for VLMs training. Code is available at https://github.com/Yu-xm/Unicorn.git.

Ming Meng, Ke Mu, Yonggui Zhu et al.

Generating expressive and diverse human gestures from audio is crucial in fields like human-computer interaction, virtual reality, and animation. Though existing methods have achieved remarkable performance, they often exhibit limitations due to constrained dataset diversity and the restricted amount of information derived from audio inputs. To address these challenges, we present VarGes, a novel variation-driven framework designed to enhance co-speech gesture generation by integrating visual stylistic cues while maintaining naturalness. Our approach begins with the Variation-Enhanced Feature Extraction (VEFE) module, which seamlessly incorporates \textcolor{blue}{style-reference} video data into a 3D human pose estimation network to extract StyleCLIPS, thereby enriching the input with stylistic information. Subsequently, we employ the Variation-Compensation Style Encoder (VCSE), a transformer-style encoder equipped with an additive attention mechanism pooling layer, to robustly encode diverse StyleCLIPS representations and effectively manage stylistic variations. Finally, the Variation-Driven Gesture Predictor (VDGP) module fuses MFCC audio features with StyleCLIPS encodings via cross-attention, injecting this fused data into a cross-conditional autoregressive model to modulate 3D human gesture generation based on audio input and stylistic clues. The efficacy of our approach is validated on benchmark datasets, where it outperforms existing methods in terms of gesture diversity and naturalness. The code and video results will be made publicly available upon acceptance:https://github.com/mookerr/VarGES/ .

Xingjian Zhang, Xi Weng, Yihao Yue et al.

Video behavior recognition and scene understanding are fundamental tasks in multimodal intelligence, serving as critical building blocks for numerous real-world applications. Through large multimodal models (LMMs) have achieved remarkable progress in video understanding, most existing open-source models rely on over 7B parameters and require large-scale datasets for training, making them resource-intensive and inaccessible to many researchers. Furthermore, lightweight models face persistent challenges in effectively processing long visual sequences and temporal understanding. In this work, we introduce TinyLLaVA-Video, a lightweight yet powerful video understanding model with approximately 3.6B parameters. The cornerstone of our design is the video-level group resampler, a novel mechanism that significantly reduces and controls the number of visual tokens at the video level. Unlike traditional image-level resampler, our approach effectively mitigates redundancy while enhancing temporal comprehension, leading to improved performance on video-based tasks. In addition, TinyLLaVA-Video demonstrates exceptional efficiency, requiring only one day of training on 8 A100-40G GPUs. It surpasses several existing 7B-parameter models on multiple benchmarks. We believe this work provides a valuable foundation for future research on lightweight video understanding models. The code and weights is available at https://github.com/ZhangXJ199/TinyLLaVA-Video.

Xingyu Wang, Pengxiang Ding, Jingkai Xu et al.

Recent advances in visuomotor policy learning have enabled robots to perform control directly from visual inputs. Yet, extending such end-to-end learning from single-arm to bimanual manipulation remains challenging due to the need for both independent perception and coordinated interaction between arms. Existing methods typically favor one side -- either decoupling the two arms to avoid interference or enforcing strong cross-arm coupling for coordination -- thus lacking a unified treatment. We propose CUBic, a Coordinated and Unified framework for Bimanual perception and control that reformulates bimanual coordination as a unified perceptual modeling problem. CUBic learns a shared tokenized representation bridging perception and control, where independence and coordination emerge intrinsically from structure rather than from hand-crafted coupling. Our approach integrates three components: unidirectional perception aggregation, bidirectional perception coordination through two codebooks with shared mapping, and a unified perception-to-control diffusion policy. Extensive experiments on the RoboTwin benchmark show that CUBic consistently surpasses standard baselines, achieving marked improvements in coordination accuracy and task success rates over state-of-the-art visuomotor baselines.

Zhaoxin Fan, Nanxiang Jiang, Daiheng Gao et al.

Removing undesired concepts from large-scale text-to-image (T2I) and text-to-video (T2V) diffusion models while preserving overall generative quality remains a major challenge, particularly as modern models such as Stable Diffusion v3, Flux, and OpenSora employ flow-matching and transformer-based architectures and extend to long-horizon video generation. Existing concept erasure methods, designed for earlier T2I/T2V models, often fail to generalize to these paradigms. To address this issue, we propose EraseAnything++, a unified framework for concept erasure in both image and video diffusion models with flow-matching objectives. Central to our approach is formulating concept erasure as a constrained multi-objective optimization problem that explicitly balances concept removal with preservation of generative utility. To solve the resulting conflicting objectives, we introduce an efficient utility-preserving unlearning strategy based on implicit gradient surgery. Furthermore, by integrating LoRA-based parameter tuning with attention-level regularization, our method anchors erasure on key visual representations and propagates it consistently across spatial and temporal dimensions. In the video setting, we further enhance consistency through an anchor-and-propagate mechanism that initializes erasure on reference frames and enforces it throughout subsequent transformer layers, thereby mitigating temporal drift. Extensive experiments on both image and video benchmarks demonstrate that EraseAnything++ substantially outperforms prior methods in erasure effectiveness, generative fidelity, and temporal consistency, establishing a new state of the art for concept erasure in next-generation diffusion models.

Kun Wang, Guibin Zhang, Zhenhong Zhou et al. · mit

The remarkable success of Large Language Models (LLMs) has illuminated a promising pathway toward achieving Artificial General Intelligence for both academic and industrial communities, owing to their unprecedented performance across various applications. As LLMs continue to gain prominence in both research and commercial domains, their security and safety implications have become a growing concern, not only for researchers and corporations but also for every nation. Currently, existing surveys on LLM safety primarily focus on specific stages of the LLM lifecycle, e.g., deployment phase or fine-tuning phase, lacking a comprehensive understanding of the entire "lifechain" of LLMs. To address this gap, this paper introduces, for the first time, the concept of "full-stack" safety to systematically consider safety issues throughout the entire process of LLM training, deployment, and eventual commercialization. Compared to the off-the-shelf LLM safety surveys, our work demonstrates several distinctive advantages: (I) Comprehensive Perspective. We define the complete LLM lifecycle as encompassing data preparation, pre-training, post-training, deployment and final commercialization. To our knowledge, this represents the first safety survey to encompass the entire lifecycle of LLMs. (II) Extensive Literature Support. Our research is grounded in an exhaustive review of over 800+ papers, ensuring comprehensive coverage and systematic organization of security issues within a more holistic understanding. (III) Unique Insights. Through systematic literature analysis, we have developed reliable roadmaps and perspectives for each chapter. Our work identifies promising research directions, including safety in data generation, alignment techniques, model editing, and LLM-based agent systems. These insights provide valuable guidance for researchers pursuing future work in this field.

Yihong Lin, Liang Peng, Zhaoxin Fan et al.

The creation of increasingly vivid 3D talking face has become a hot topic in recent years. Currently, most speech-driven works focus on lip synchronisation but neglect to effectively capture the correlations between emotions and facial motions. To address this problem, we propose a two-stream network called EmoFace, which consists of an emotion branch and a content branch. EmoFace employs a novel Mesh Attention mechanism to analyse and fuse the emotion features and content features. Particularly, a newly designed spatio-temporal graph-based convolution, SpiralConv3D, is used in Mesh Attention to learn potential temporal and spatial feature dependencies between mesh vertices. In addition, to the best of our knowledge, it is the first time to introduce a new self-growing training scheme with intermediate supervision to dynamically adjust the ratio of groundtruth adopted in the 3D face animation task. Comprehensive quantitative and qualitative evaluations on our high-quality 3D emotional facial animation dataset, 3D-RAVDESS ($4.8863\times 10^{-5}$mm for LVE and $0.9509\times 10^{-5}$mm for EVE), together with the public dataset VOCASET ($2.8669\times 10^{-5}$mm for LVE and $0.4664\times 10^{-5}$mm for EVE), demonstrate that our approach achieves state-of-the-art performance.

Daiheng Gao, Shilin Lu, Shaw Walters et al.

Removing unwanted concepts from large-scale text-to-image (T2I) diffusion models while maintaining their overall generative quality remains an open challenge. This difficulty is especially pronounced in emerging paradigms, such as Stable Diffusion (SD) v3 and Flux, which incorporate flow matching and transformer-based architectures. These advancements limit the transferability of existing concept-erasure techniques that were originally designed for the previous T2I paradigm (e.g., SD v1.4). In this work, we introduce EraseAnything, the first method specifically developed to address concept erasure within the latest flow-based T2I framework. We formulate concept erasure as a bi-level optimization problem, employing LoRA-based parameter tuning and an attention map regularizer to selectively suppress undesirable activations. Furthermore, we propose a self-contrastive learning strategy to ensure that removing unwanted concepts does not inadvertently harm performance on unrelated ones. Experimental results demonstrate that EraseAnything successfully fills the research gap left by earlier methods in this new T2I paradigm, achieving state-of-the-art performance across a wide range of concept erasure tasks.

Zhiyu Li, Shichao Song, Chenyang Xi et al.

Large Language Models (LLMs) have become an essential infrastructure for Artificial General Intelligence (AGI), yet their lack of well-defined memory management systems hinders the development of long-context reasoning, continual personalization, and knowledge consistency.Existing models mainly rely on static parameters and short-lived contextual states, limiting their ability to track user preferences or update knowledge over extended periods.While Retrieval-Augmented Generation (RAG) introduces external knowledge in plain text, it remains a stateless workaround without lifecycle control or integration with persistent representations.Recent work has modeled the training and inference cost of LLMs from a memory hierarchy perspective, showing that introducing an explicit memory layer between parameter memory and external retrieval can substantially reduce these costs by externalizing specific knowledge. Beyond computational efficiency, LLMs face broader challenges arising from how information is distributed over time and context, requiring systems capable of managing heterogeneous knowledge spanning different temporal scales and sources. To address this challenge, we propose MemOS, a memory operating system that treats memory as a manageable system resource. It unifies the representation, scheduling, and evolution of plaintext, activation-based, and parameter-level memories, enabling cost-efficient storage and retrieval. As the basic unit, a MemCube encapsulates both memory content and metadata such as provenance and versioning. MemCubes can be composed, migrated, and fused over time, enabling flexible transitions between memory types and bridging retrieval with parameter-based learning. MemOS establishes a memory-centric system framework that brings controllability, plasticity, and evolvability to LLMs, laying the foundation for continual learning and personalized modeling.

Zhefei Gong, Pengxiang Ding, Shangke Lyu et al.

In robotic visuomotor policy learning, diffusion-based models have achieved significant success in improving the accuracy of action trajectory generation compared to traditional autoregressive models. However, they suffer from inefficiency due to multiple denoising steps and limited flexibility from complex constraints. In this paper, we introduce Coarse-to-Fine AutoRegressive Policy (CARP), a novel paradigm for visuomotor policy learning that redefines the autoregressive action generation process as a coarse-to-fine, next-scale approach. CARP decouples action generation into two stages: first, an action autoencoder learns multi-scale representations of the entire action sequence; then, a GPT-style transformer refines the sequence prediction through a coarse-to-fine autoregressive process. This straightforward and intuitive approach produces highly accurate and smooth actions, matching or even surpassing the performance of diffusion-based policies while maintaining efficiency on par with autoregressive policies. We conduct extensive evaluations across diverse settings, including single-task and multi-task scenarios on state-based and image-based simulation benchmarks, as well as real-world tasks. CARP achieves competitive success rates, with up to a 10% improvement, and delivers 10x faster inference compared to state-of-the-art policies, establishing a high-performance, efficient, and flexible paradigm for action generation in robotic tasks.

Bozhi Luan, Gen Li, Yalan Qin et al.

We address hallucination detection in Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) by framing the problem through the lens of dynamical systems stability theory. Rather than treating hallucination as a straightforward classification task, we conceptualize (M)LLMs as dynamical systems, where factual knowledge is represented by stable equilibrium points within the representation space. Our main insight is that hallucinations tend to arise at the boundaries of knowledge-transition regions separating stable and unstable zones. To capture this phenomenon, we propose Lyapunov Probes: lightweight networks trained with derivative-based stability constraints that enforce a monotonic decay in confidence under input perturbations. By performing systematic perturbation analysis and applying a two-stage training process, these probes reliably distinguish between stable factual regions and unstable, hallucination-prone regions. Experiments on diverse datasets and models demonstrate consistent improvements over existing baselines.

Mingjin Chen, Junhao Chen, Xiaojun Ye et al.

3D human body reconstruction has been a challenge in the field of computer vision. Previous methods are often time-consuming and difficult to capture the detailed appearance of the human body. In this paper, we propose a new method called \emph{Ultraman} for fast reconstruction of textured 3D human models from a single image. Compared to existing techniques, \emph{Ultraman} greatly improves the reconstruction speed and accuracy while preserving high-quality texture details. We present a set of new frameworks for human reconstruction consisting of three parts, geometric reconstruction, texture generation and texture mapping. Firstly, a mesh reconstruction framework is used, which accurately extracts 3D human shapes from a single image. At the same time, we propose a method to generate a multi-view consistent image of the human body based on a single image. This is finally combined with a novel texture mapping method to optimize texture details and ensure color consistency during reconstruction. Through extensive experiments and evaluations, we demonstrate the superior performance of \emph{Ultraman} on various standard datasets. In addition, \emph{Ultraman} outperforms state-of-the-art methods in terms of human rendering quality and speed. Upon acceptance of the article, we will make the code and data publicly available.

Yuanze Hu, Gen Li, Yuqin Lan et al.

Multimodal large language models (MLLMs) have achieved impressive progress on general multimodal tasks, yet they remain brittle on dial-based measurement reading. In this paper, we study this problem through controlled benchmarks and feature-space probing, and show that current MLLMs not only achieve unsatisfactory accuracy on dial-based readout, but also suffer sharp performance drops under viewpoint and illumination changes even when the underlying dial state remains fixed. Our probing analysis further reveals that same-state samples under appearance variation are not consistently clustered, while neighboring states fail to preserve the local structure implied by continuous dial values. These findings suggest that existing MLLMs largely ignore the intrinsic state geometry of dial measurement tasks and instead rely on superficial appearance cues. Motivated by this diagnosis, we propose TriSCA, a tri-level state-consistent alignment framework for dial-based measurement reading. Specifically, TriSCA consists of state-distance-aware representation alignment, metadata-grounded observation-to-state supervision, and state-aware objective alignment. Extensive ablation studies and evaluation experiments on controlled clock and gauge benchmarks, together with evaluation on an external real-world benchmark, demonstrate the effectiveness of our method.

Xudong Cai, Yongcai Wang, Zhaoxin Fan et al.

Photo-realistic scene reconstruction from sparse-view, uncalibrated images is highly required in practice. Although some successes have been made, existing methods are either Sparse-View but require accurate camera parameters (i.e., intrinsic and extrinsic), or SfM-free but need densely captured images. To combine the advantages of both methods while addressing their respective weaknesses, we propose Dust to Tower (D2T), an accurate and efficient coarse-to-fine framework to optimize 3DGS and image poses simultaneously from sparse and uncalibrated images. Our key idea is to first construct a coarse model efficiently and subsequently refine it using warped and inpainted images at novel viewpoints. To do this, we first introduce a Coarse Construction Module (CCM) which exploits a fast Multi-View Stereo model to initialize a 3D Gaussian Splatting (3DGS) and recover initial camera poses. To refine the 3D model at novel viewpoints, we propose a Confidence Aware Depth Alignment (CADA) module to refine the coarse depth maps by aligning their confident parts with estimated depths by a Mono-depth model. Then, a Warped Image-Guided Inpainting (WIGI) module is proposed to warp the training images to novel viewpoints by the refined depth maps, and inpainting is applied to fulfill the ``holes" in the warped images caused by view-direction changes, providing high-quality supervision to further optimize the 3D model and the camera poses. Extensive experiments and ablation studies demonstrate the validity of D2T and its design choices, achieving state-of-the-art performance in both tasks of novel view synthesis and pose estimation while keeping high efficiency. Codes will be publicly available.

Ziqiao Peng, Yanbo Fan, Haoyu Wu et al.

In face-to-face conversations, individuals need to switch between speaking and listening roles seamlessly. Existing 3D talking head generation models focus solely on speaking or listening, neglecting the natural dynamics of interactive conversation, which leads to unnatural interactions and awkward transitions. To address this issue, we propose a new task -- multi-round dual-speaker interaction for 3D talking head generation -- which requires models to handle and generate both speaking and listening behaviors in continuous conversation. To solve this task, we introduce DualTalk, a novel unified framework that integrates the dynamic behaviors of speakers and listeners to simulate realistic and coherent dialogue interactions. This framework not only synthesizes lifelike talking heads when speaking but also generates continuous and vivid non-verbal feedback when listening, effectively capturing the interplay between the roles. We also create a new dataset featuring 50 hours of multi-round conversations with over 1,000 characters, where participants continuously switch between speaking and listening roles. Extensive experiments demonstrate that our method significantly enhances the naturalness and expressiveness of 3D talking heads in dual-speaker conversations. We recommend watching the supplementary video: https://ziqiaopeng.github.io/dualtalk.

Feiyuan Zhang, Dezhi Zhu, James Ming et al.

Retrieval-Augmented Generation (RAG) systems have shown substantial benefits in applications such as question answering and multi-turn dialogue \citep{lewis2020retrieval}. However, traditional RAG methods, while leveraging static knowledge bases, often overlook the potential of dynamic historical information in ongoing conversations. To bridge this gap, we introduce DH-RAG, a Dynamic Historical Context-Powered Retrieval-Augmented Generation Method for Multi-Turn Dialogue. DH-RAG is inspired by human cognitive processes that utilize both long-term memory and immediate historical context in conversational responses \citep{stafford1987conversational}. DH-RAG is structured around two principal components: a History-Learning based Query Reconstruction Module, designed to generate effective queries by synthesizing current and prior interactions, and a Dynamic History Information Updating Module, which continually refreshes historical context throughout the dialogue. The center of DH-RAG is a Dynamic Historical Information database, which is further refined by three strategies within the Query Reconstruction Module: Historical Query Clustering, Hierarchical Matching, and Chain of Thought Tracking. Experimental evaluations show that DH-RAG significantly surpasses conventional models on several benchmarks, enhancing response relevance, coherence, and dialogue quality.

Bofang Jia, Pengxiang Ding, Can Cui et al.

Visual-motor policy learning has advanced with architectures like diffusion-based policies, known for modeling complex robotic trajectories. However, their prolonged inference times hinder high-frequency control tasks requiring real-time feedback. While consistency distillation (CD) accelerates inference, it introduces errors that compromise action quality. To address these limitations, we propose the Score and Distribution Matching Policy (SDM Policy), which transforms diffusion-based policies into single-step generators through a two-stage optimization process: score matching ensures alignment with true action distributions, and distribution matching minimizes KL divergence for consistency. A dual-teacher mechanism integrates a frozen teacher for stability and an unfrozen teacher for adversarial training, enhancing robustness and alignment with target distributions. Evaluated on a 57-task simulation benchmark, SDM Policy achieves a 6x inference speedup while having state-of-the-art action quality, providing an efficient and reliable framework for high-frequency robotic tasks.

Junhao Chen, Xiang Li, Xiaojun Ye et al.

With the success of 2D diffusion models, 2D AIGC content has already transformed our lives. Recently, this success has been extended to 3D AIGC, with state-of-the-art methods generating textured 3D models from single images or text. However, we argue that current 3D AIGC methods still do not fully unleash human creativity. We often imagine 3D content made from multimodal inputs, such as what it would look like if my pet bunny were eating a doughnut on the table. In this paper, we explore a novel 3D AIGC approach: generating 3D content from IDEAs. An IDEA is a multimodal input composed of text, image, and 3D models. To our knowledge, this challenging and exciting 3D AIGC setting has not been studied before. We propose the new framework Idea23D, which combines three agents based on large multimodal models (LMMs) and existing algorithmic tools. These three LMM-based agents are tasked with prompt generation, model selection, and feedback reflection. They collaborate and critique each other in a fully automated loop, without human intervention. The framework then generates a text prompt to create 3D models that align closely with the input IDEAs. We demonstrate impressive 3D AIGC results that surpass previous methods. To comprehensively assess the 3D AIGC capabilities of Idea23D, we introduce the Eval3DAIGC-198 dataset, containing 198 multimodal inputs for 3D generation tasks. This dataset evaluates the alignment between generated 3D content and input IDEAs. Our user study and quantitative results show that Idea23D significantly improves the success rate and accuracy of 3D generation, with excellent compatibility across various LMM, Text-to-Image, and Image-to-3D models. Code and dataset are available at \url{https://idea23d.github.io/}.

Jihao Zhao, Zhiyuan Ji, Zhaoxin Fan et al.

Retrieval-Augmented Generation (RAG), while serving as a viable complement to large language models (LLMs), often overlooks the crucial aspect of text chunking within its pipeline. This paper initially introduces a dual-metric evaluation method, comprising Boundary Clarity and Chunk Stickiness, to enable the direct quantification of chunking quality. Leveraging this assessment method, we highlight the inherent limitations of traditional and semantic chunking in handling complex contextual nuances, thereby substantiating the necessity of integrating LLMs into chunking process. To address the inherent trade-off between computational efficiency and chunking precision in LLM-based approaches, we devise the granularity-aware Mixture-of-Chunkers (MoC) framework, which consists of a three-stage processing mechanism. Notably, our objective is to guide the chunker towards generating a structured list of chunking regular expressions, which are subsequently employed to extract chunks from the original text. Extensive experiments demonstrate that both our proposed metrics and the MoC framework effectively settle challenges of the chunking task, revealing the chunking kernel while enhancing the performance of the RAG system.

Kaixing Yang, Xulong Tang, Ziqiao Peng et al.

Music-to-dance generation represents a challenging yet pivotal task at the intersection of choreography, virtual reality, and creative content generation. Despite its significance, existing methods face substantial limitation in achieving choreographic consistency. To address the challenge, we propose MatchDance, a novel framework for music-to-dance generation that constructs a latent representation to enhance choreographic consistency. MatchDance employs a two-stage design: (1) a Kinematic-Dynamic-based Quantization Stage (KDQS), which encodes dance motions into a latent representation by Finite Scalar Quantization (FSQ) with kinematic-dynamic constraints and reconstructs them with high fidelity, and (2) a Hybrid Music-to-Dance Generation Stage(HMDGS), which uses a Mamba-Transformer hybrid architecture to map music into the latent representation, followed by the KDQS decoder to generate 3D dance motions. Additionally, a music-dance retrieval framework and comprehensive metrics are introduced for evaluation. Extensive experiments on the FineDance dataset demonstrate state-of-the-art performance.