Yuhan Zhu

Guozhen Zhang, Yuhan Zhu, Haonan Wang et al.

Effectively extracting inter-frame motion and appearance information is important for video frame interpolation (VFI). Previous works either extract both types of information in a mixed way or elaborate separate modules for each type of information, which lead to representation ambiguity and low efficiency. In this paper, we propose a novel module to explicitly extract motion and appearance information via a unifying operation. Specifically, we rethink the information process in inter-frame attention and reuse its attention map for both appearance feature enhancement and motion information extraction. Furthermore, for efficient VFI, our proposed module could be seamlessly integrated into a hybrid CNN and Transformer architecture. This hybrid pipeline can alleviate the computational complexity of inter-frame attention as well as preserve detailed low-level structure information. Experimental results demonstrate that, for both fixed- and arbitrary-timestep interpolation, our method achieves state-of-the-art performance on various datasets. Meanwhile, our approach enjoys a lighter computation overhead over models with close performance. The source code and models are available at https://github.com/MCG-NJU/EMA-VFI.

Chen Xu, Yuhan Zhu, Haocheng Shen et al.

Prompt learning has been designed as an alternative to fine-tuning for adapting Vision-language (V-L) models to the downstream tasks. Previous works mainly focus on text prompt while visual prompt works are limited for V-L models. The existing visual prompt methods endure either mediocre performance or unstable training process, indicating the difficulty of visual prompt learning. In this paper, we propose a new Progressive Visual Prompt (ProVP) structure to strengthen the interactions among prompts of different layers. More importantly, our ProVP could effectively propagate the image embeddings to deep layers and behave partially similar to an instance adaptive prompt method. To alleviate generalization deterioration, we further propose a new contrastive feature re-formation, which prevents the serious deviation of the prompted visual feature from the fixed CLIP visual feature distribution. Combining both, our method (ProVP-Ref) is evaluated on 11 image benchmark datasets and achieves 7/11 state-of-theart results on both few-shot and base-to-novel settings. To the best of our knowledge, we are the first to demonstrate the superior performance of visual prompts in V-L models to previous prompt-based methods in downstream tasks. Meanwhile, it implies that our ProVP-Ref shows the best capability to adapt and to generalize.

Chen Xu, Yuhan Zhu, Guozhen Zhang et al.

Prompt learning has emerged as an efficient and effective approach for transferring foundational Vision-Language Models (e.g., CLIP) to downstream tasks. However, current methods tend to overfit to seen categories, thereby limiting their generalization ability for unseen classes. In this paper, we propose a new method, Decoupled Prompt Learning (DPL), which reformulates the attention in prompt learning to alleviate this problem. Specifically, we theoretically investigate the collaborative process between prompts and instances (i.e., image patches/text tokens) by reformulating the original self-attention into four separate sub-processes. Through detailed analysis, we observe that certain sub-processes can be strengthened to bolster robustness and generalizability by some approximation techniques. Furthermore, we introduce language-conditioned textual prompting based on decoupled attention to naturally preserve the generalization of text input. Our approach is flexible for both visual and textual modalities, making it easily extendable to multi-modal prompt learning. By combining the proposed techniques, our approach achieves state-of-the-art performance on three representative benchmarks encompassing 15 image recognition datasets, while maintaining parameter-efficient. Moreover, our DPL does not rely on any auxiliary regularization task or extra training data, further demonstrating its remarkable generalization ability.

Xinhao Li, Yuhan Zhu, Limin Wang

Adapting image models to the video domain has emerged as an efficient paradigm for solving video recognition tasks. Due to the huge number of parameters and effective transferability of image models, performing full fine-tuning is less efficient and even unnecessary. Thus, recent research is shifting its focus toward parameter-efficient image-to-video adaptation. However, these adaptation strategies inevitably introduce extra computational costs to deal with the domain gap and temporal modeling in videos. In this paper, we present a new adaptation paradigm (ZeroI2V) to transfer the image transformers to video recognition tasks (i.e., introduce zero extra cost to the original models during inference). To achieve this goal, we present two core designs. First, to capture the dynamics in videos and reduce the difficulty of image-to-video adaptation, we exploit the flexibility of self-attention and introduce spatial-temporal dual-headed attention (STDHA). This approach efficiently endows the image transformers with temporal modeling capability at zero extra parameters and computation. Second, to handle the domain gap between images and videos, we propose a linear adaption strategy that utilizes lightweight densely placed linear adapters to fully transfer the frozen image models to video recognition. Thanks to the customized linear design, all newly added adapters could be easily merged with the original modules through structural reparameterization after training, enabling zero extra cost during inference. Extensive experiments on representative fully-supervised and few-shot video recognition benchmarks showcase that ZeroI2V can match or even outperform previous state-of-the-art methods while enjoying superior parameter and inference efficiency.

Yuhan Zhu, Yuyang Ji, Zhiyu Zhao et al.

Pre-trained vision-language models (VLMs) have shown impressive results in various visual classification tasks. However, we often fail to fully unleash their potential when adapting them for new concept understanding due to limited information on new classes. To address this limitation, we introduce a novel adaptation framework, AWT (Augment, Weight, then Transport). AWT comprises three key components: augmenting inputs with diverse visual perspectives and enriched class descriptions through image transformations and language models; dynamically weighting inputs based on the prediction entropy; and employing optimal transport to mine semantic correlations in the vision-language space. AWT can be seamlessly integrated into various VLMs, enhancing their zero-shot capabilities without additional training and facilitating few-shot learning through an integrated multimodal adapter module. We verify AWT in multiple challenging scenarios, including zero-shot and few-shot image classification, zero-shot video action recognition, and out-of-distribution generalization. AWT consistently outperforms the state-of-the-art methods in each setting. In addition, our extensive studies further demonstrate AWT's effectiveness and adaptability across different VLMs, architectures, and scales.

Xiangyu Zeng, Zhiqiu Zhang, Yuhan Zhu et al.

Existing multimodal large language models for long-video understanding predominantly rely on uniform sampling and single-turn inference, limiting their ability to identify sparse yet critical evidence amid extensive redundancy. We introduce Video-o3, a novel framework that supports iterative discovery of salient visual clues, fine-grained inspection of key segments, and adaptive termination once sufficient evidence is acquired. Technically, we address two core challenges in interleaved tool invocation. First, to mitigate attention dispersion induced by the heterogeneity of reasoning and tool-calling, we propose Task-Decoupled Attention Masking, which isolates per-step concentration while preserving shared global context. Second, to control context length growth in multi-turn interactions, we introduce a Verifiable Trajectory-Guided Reward that balances exploration coverage with reasoning efficiency. To support training at scale, we further develop a data synthesis pipeline and construct Seeker-173K, comprising 173K high-quality tool-interaction trajectories for effective supervised and reinforcement learning. Extensive experiments show that Video-o3 substantially outperforms state-of-the-art methods, achieving 72.1% accuracy on MLVU and 46.5% on Video-Holmes. These results demonstrate Video-o3's strong multi-hop evidence-seeking and reasoning capabilities, and validate the effectiveness of native tool invocation in long-video scenarios.

Xinhao Li, Yi Wang, Jiashuo Yu et al.

Long-context video modeling is critical for multimodal large language models (MLLMs), enabling them to process movies, online video streams, and so on. Despite its advances, handling long videos remains challenging due to the difficulty in efficiently understanding the extremely long video context. This paper aims to address this issue from aspects of model architecture, training data, training strategy and evaluation benchmark. First, we propose a novel Hierarchical video token Compression (HiCo) method, which leverages visual redundancy in long videos to compress long video context from Clip-level to Video-level, reducing the computation significantly while preserving essential details, achieving an extreme compression ratio of approximately 1/50 with almost no performance loss. Second, we introduce a multi-stage short-to-long learning scheme, a large-scale dataset of real-world long videos named LongVid, and a challenging ``Multi-Hop Needle-In-A-Video-Haystack'' benchmark. Finally, we build a powerful video MLLM named VideoChat-Flash, which shows a leading performance on both mainstream long and short video benchmarks at the 2B and 7B model scale. It first gets 99.1% accuracy over 10,000 frames in NIAH among open-source models.

Yuhan Zhu, Guozhen Zhang, Chen Xu et al.

Vision-language models have showcased impressive zero-shot classification capabilities when equipped with suitable text prompts. Previous studies have shown the effectiveness of test-time prompt tuning; however, these methods typically require per-image prompt adaptation during inference, which incurs high computational budgets and limits scalability and practical deployment. To overcome this issue, we introduce Self-TPT, a novel framework leveraging Self-supervised learning for efficient Test-time Prompt Tuning. The key aspect of Self-TPT is that it turns to efficient predefined class adaptation via self-supervised learning, thus avoiding computation-heavy per-image adaptation at inference. Self-TPT begins by co-training the self-supervised and the classification task using source data, then applies the self-supervised task exclusively for test-time new class adaptation. Specifically, we propose Contrastive Prompt Learning (CPT) as the key task for self-supervision. CPT is designed to minimize the intra-class distances while enhancing inter-class distinguishability via contrastive learning. Furthermore, empirical evidence suggests that CPT could closely mimic back-propagated gradients of the classification task, offering a plausible explanation for its effectiveness. Motivated by this finding, we further introduce a gradient matching loss to explicitly enhance the gradient similarity. We evaluated Self-TPT across three challenging zero-shot benchmarks. The results consistently demonstrate that Self-TPT not only significantly reduces inference costs but also achieves state-of-the-art performance, effectively balancing the efficiency-efficacy trade-off.

Chenting Wang, Yuhan Zhu, Yicheng Xu et al.

Large-scale video-text pretraining achieves strong performance but depends on noisy, synthetic captions with limited semantic coverage, often overlooking implicit world knowledge such as object motion, 3D geometry, and physical cues. In contrast, masked video modeling (MVM) directly exploits spatiotemporal structures but trails text-supervised methods on general tasks. We find this gap arises from overlooked architectural issues: pixel-level reconstruction struggles with convergence and its low-level requirement often conflicts with semantics, while latent prediction often encourages shortcut learning. To address these, we disentangle the traditional encoder-decoder design into an Encoder-Predictor-Decoder (EPD) framework, where the predictor acts as a latent world model, and propose InternVideo-Next, a two-stage pretraining scheme that builds a semantically consistent yet detail-preserving latent space for this world model. First, conventional linear decoder in pixel MVM enforces the predictor output latent to be linearly projected to, thus separable in pixel space, causing the conflict with semantic abstraction. Our Stage 1 proposes a conditional diffusion decoder and injects reliable image-level semantic priors to enhance semantics and convergence, thus bridging pixel-level fidelity with high-level semantic abstraction. Stage 2 further learns world knowledge by predicting frozen Stage 1 targets within this space, mitigating shortcut learning. Trained on public, unlabeled videos, InternVideo-Next achieves state-of-the-art results across benchmarks and provides a scalable path toward general video representation learning.

Yuhan Zhu, Yanyu Zhang, Jie Xu et al.

3D Gaussian Splatting (3DGS) has shown promising results for 3D scene modeling using mixtures of Gaussians, yet its existing simultaneous localization and mapping (SLAM) variants typically rely on direct, deterministic pose optimization against the splat map, making them sensitive to initialization and susceptible to catastrophic forgetting as map evolves. We propose Variational Bayesian Gaussian Splatting SLAM (VBGS-SLAM), a novel framework that couples the splat map refinement and camera pose tracking in a generative probabilistic form. By leveraging conjugate properties of multivariate Gaussians and variational inference, our method admits efficient closed-form updates and explicitly maintains posterior uncertainty over both poses and scene parameters. This uncertainty-aware method mitigates drift and enhances robustness in challenging conditions, while preserving the efficiency and rendering quality of existing 3DGS. Our experiments demonstrate superior tracking performance and robustness in long sequence prediction, alongside efficient, high-quality novel view synthesis across diverse synthetic and real-world scenes.

Yuhan Zhu, Guozhen Zhang, Jing Tan et al.

Temporal Action Detection (TAD) aims to identify the action boundaries and the corresponding category within untrimmed videos. Inspired by the success of DETR in object detection, several methods have adapted the query-based framework to the TAD task. However, these approaches primarily followed DETR to predict actions at the instance level (i.e., identify each action by its center point), leading to sub-optimal boundary localization. To address this issue, we propose a new Dual-level query-based TAD framework, namely DualDETR, to detect actions from both instance-level and boundary-level. Decoding at different levels requires semantics of different granularity, therefore we introduce a two-branch decoding structure. This structure builds distinctive decoding processes for different levels, facilitating explicit capture of temporal cues and semantics at each level. On top of the two-branch design, we present a joint query initialization strategy to align queries from both levels. Specifically, we leverage encoder proposals to match queries from each level in a one-to-one manner. Then, the matched queries are initialized using position and content prior from the matched action proposal. The aligned dual-level queries can refine the matched proposal with complementary cues during subsequent decoding. We evaluate DualDETR on three challenging multi-label TAD benchmarks. The experimental results demonstrate the superior performance of DualDETR to the existing state-of-the-art methods, achieving a substantial improvement under det-mAP and delivering impressive results under seg-mAP.

Xiangfei Qiu, Yuhan Zhu, Zhengyu Li et al.

Time series forecasting is crucial in various fields such as economics, traffic, and AIOps. However, in real-world applications, focusing solely on the endogenous variables (i.e., target variables), is often insufficient to ensure accurate predictions. Considering exogenous variables (i.e., covariates) provides additional predictive information, thereby improving forecasting accuracy. However, existing methods for time series forecasting with exogenous variables (TSF-X) have the following shortcomings: 1) they do not leverage future exogenous variables, 2) they fail to account for the causal relationships between endogenous and exogenous variables. As a result, their performance is suboptimal. In this study, to better leverage exogenous variables, especially future exogenous variable, we propose a general framework DAG, which utilizes dual causal network along both the temporal and channel dimensions for time series forecasting with exogenous variables. Specifically, we first introduce the Temporal Causal Module, which includes a causal discovery module to capture how historical exogenous variables affect future exogenous variables. Following this, we construct a causal injection module that incorporates the discovered causal relationships into the process of forecasting future endogenous variables based on historical endogenous variables. Next, we propose the Channel Causal Module, which follows a similar design principle. It features a causal discovery module models how historical exogenous variables influence historical endogenous variables, and a causal injection module incorporates the discovered relationships to enhance the prediction of future endogenous variables based on future exogenous variables.

Guozhen Zhang, Yuhan Zhu, Yutao Cui et al.

Flow-based frame interpolation methods ensure motion stability through estimated intermediate flow but often introduce severe artifacts in complex motion regions. Recent generative approaches, boosted by large-scale pre-trained video generation models, show promise in handling intricate scenes. However, they frequently produce unstable motion and content inconsistencies due to the absence of explicit motion trajectory constraints. To address these challenges, we propose Motion-aware Generative frame interpolation (MoG) that synergizes intermediate flow guidance with generative capacities to enhance interpolation fidelity. Our key insight is to simultaneously enforce motion smoothness through flow constraints while adaptively correcting flow estimation errors through generative refinement. Specifically, we first introduce a dual guidance injection that propagates condition information using intermediate flow at both latent and feature levels, aligning the generated motion with flow-derived motion trajectories. Meanwhile, we implemented two critical designs, encoder-only guidance injection and selective parameter fine-tuning, which enable dynamic artifact correction in the complex motion regions. Extensive experiments on both real-world and animation benchmarks demonstrate that MoG outperforms state-of-the-art methods in terms of video quality and visual fidelity. Our work bridges the gap between flow-based stability and generative flexibility, offering a versatile solution for frame interpolation across diverse scenarios.

Zhengyu Li, Xiangfei Qiu, Yuhan Zhu et al.

Exogenous variables offer valuable supplementary information for predicting future endogenous variables. Forecasting with exogenous variables needs to consider both past-to-future dependencies (i.e., temporal correlations) and the influence of exogenous variables on endogenous variables (i.e., channel correlations). This is pivotal when future exogenous variables are available, because they may directly affect the future endogenous variables. Many methods have been proposed for time series forecasting with exogenous variables, focusing on modeling temporal and channel correlations. However, most of them use a two-step strategy, modeling temporal and channel correlations separately, which limits their ability to capture joint correlations across time and channels. Furthermore, in real-world scenarios, time series are frequently affected by various forms of noises, underscoring the critical importance of robustness in such correlations modeling. To address these limitations, we propose GCGNet, a Graph-Consistent Generative Network for time series forecasting with exogenous variables. Specifically, GCGNet first employs a Variational Generator to produce coarse predictions. A Graph Structure Aligner then further guides it by evaluating the consistency between the generated and true correlations, where the correlations are represented as graphs, and are robust to noises. Finally, a Graph Refiner is proposed to refine the predictions to prevent degeneration and improve accuracy. Extensive experiments on 12 real-world datasets demonstrate that GCGNet outperforms state-of-the-art baselines.

Chunxu Liu, Jiyuan Yang, Ruopeng Gao et al.

Multimodal embeddings are widely used in downstream tasks such as multimodal retrieval, enabling alignment of interleaved modalities in a shared representation space. While recent studies show that Multimodal Large Language Models (MLLMs) can serve as strong embedding extractors, existing approaches treat embedding extraction as a direct encoding step, overlooking the fact that MLLMs possess the generative capability for reasoning that could be leveraged to enhance representation quality. In this work, we explore how to explicitly incorporate reasoning into the embedding process. To this end, we propose Reasoning Guided Embeddings (RGE), which preserves the generative rationale process of MLLMs and couples it with contrastive training. Our method first enables the model to perform structured rationale generation conditioned on the instruction, and then extracts representations after reasoning has unfolded. This simple design enhances the context-conditional inference signals within the embedding, leading to improved multimodal representation quality. Experiments on the MMEB benchmark show that reasoning-guided conditioning improves multimodal retrieval performance by 4.9% over the non-reasoning baseline, confirming that explicit reasoning can effectively enhance embedding quality.

Yuhan Zhu, Xiangyu Zeng, Chenting Wang et al.

Multimodal large language models (MLLMs) are emerging as versatile foundations for mixed-modality retrieval. Yet, they often require heavy post-hoc training to convert them into contrastive encoders for retrieval. This work asks: Can off-the-shelf MLLMs serve as powerful retrievers without additional training? We present FreeRet, a plug-and-play framework that turns any MLLM into a two-stage retriever. FreeRet first derives semantically grounded embeddings directly from the model for fast candidate search, and then exploits its reasoning ability for precise reranking. The framework contributes three advances: bypassing lexical alignment layers to obtain semantically faithful embeddings, conditioning representation generation with explicit priors, and mitigating framing effect in reranking via neutral choice framing. On the MMEB and MMEB-V2 benchmarks spanning 46 datasets, FreeRet substantially outperforms models trained on millions of pairs. Beyond benchmarks, FreeRet is model-agnostic and scales seamlessly across MLLM families and sizes, preserves their generative abilities, supports arbitrary modality combinations, and unifies retrieval, reranking, and generation into end-to-end RAG within a single model. Our findings demonstrate that pretrained MLLMs, when carefully harnessed, can serve as strong retrieval engines without training, closing a critical gap in their role as generalists.

Yuhan Zhu, Haojie Liu, Jian Wang et al.

The rise of large model-based AI agents has spurred interest in Multi-Agent Systems (MAS) for their capabilities in decision-making, collaboration, and adaptability. While the Model Context Protocol (MCP) addresses tool invocation and data exchange challenges via a unified protocol, it lacks support for organizing agent-level collaboration. To bridge this gap, we propose Agent-as-a-Service based on Agent Network (AaaS-AN), a service-oriented paradigm grounded in the Role-Goal-Process-Service (RGPS) standard. AaaS-AN unifies the entire agent lifecycle, including construction, integration, interoperability, and networked collaboration, through two core components: (1) a dynamic Agent Network, which models agents and agent groups as vertexes that self-organize within the network based on task and role dependencies; (2) service-oriented agents, incorporating service discovery, registration, and interoperability protocols. These are orchestrated by a Service Scheduler, which leverages an Execution Graph to enable distributed coordination, context tracking, and runtime task management. We validate AaaS-AN on mathematical reasoning and application-level code generation tasks, which outperforms state-of-the-art baselines. Notably, we constructed a MAS based on AaaS-AN containing agent groups, Robotic Process Automation (RPA) workflows, and MCP servers over 100 agent services. We also release a dataset containing 10,000 long-horizon multi-agent workflows to facilitate future research on long-chain collaboration in MAS.

Minrui Chen, Yi Zhou, Huidong Jiang et al.

Fever of unknown origin FUO remains a diagnostic challenge. MedMimic is introduced as a multimodal framework inspired by real-world diagnostic processes. It uses pretrained models such as DINOv2, Vision Transformer, and ResNet-18 to convert high-dimensional 18F-FDG PET/CT imaging into low-dimensional, semantically meaningful features. A learnable self-attention-based fusion network then integrates these imaging features with clinical data for classification. Using 416 FUO patient cases from Sichuan University West China Hospital from 2017 to 2023, the multimodal fusion classification network MFCN achieved macro-AUROC scores ranging from 0.8654 to 0.9291 across seven tasks, outperforming conventional machine learning and single-modality deep learning methods. Ablation studies and five-fold cross-validation further validated its effectiveness. By combining the strengths of pretrained large models and deep learning, MedMimic offers a promising solution for disease classification.

Yuhan Zhu, Jian Wang, Bing Li et al.

With the development of cloud-native technologies, microservice-based software systems face challenges in accurately localizing root causes when failures occur. Additionally, the cloud-edge collaborative environment introduces more difficulties, such as unstable networks and high latency across network segments. Accurately identifying the root cause of microservices in a cloud-edge collaborative environment has thus become an urgent problem. In this paper, we propose MicroCERCL, a novel approach that pinpoints root causes at the kernel and application level in the cloud-edge collaborative environment. Our key insight is that failures propagate through direct invocations and indirect resource-competition dependencies in a cloud-edge collaborative environment characterized by instability and high latency. This will become more complex in the hybrid deployment that simultaneously involves multiple microservice systems. Leveraging this insight, we extract valid contents from kernel-level logs to prioritize localizing the kernel-level root cause. Moreover, we construct a heterogeneous dynamic topology stack and train a graph neural network model to accurately localize the application-level root cause without relying on historical data. Notably, we released the first benchmark hybrid deployment microservice system in a cloud-edge collaborative environment (the largest and most complex within our knowledge). Experiments conducted on the dataset collected from the benchmark show that MicroCERCL can accurately localize the root cause of microservice systems in such environments, significantly outperforming state-of-the-art approaches with an increase of at least 24.1% in top-1 accuracy.

Aaron Hao Tan, Federico Pizarro Bejarano, Yuhan Zhu et al.

Cooperative multi-robot teams need to be able to explore cluttered and unstructured environments while dealing with communication dropouts that prevent them from exchanging local information to maintain team coordination. Therefore, robots need to consider high-level teammate intentions during action selection. In this letter, we present the first Macro Action Decentralized Exploration Network (MADE-Net) using multi-agent deep reinforcement learning (DRL) to address the challenges of communication dropouts during multi-robot exploration in unseen, unstructured, and cluttered environments. Simulated robot team exploration experiments were conducted and compared against classical and DRL methods where MADE-Net outperformed all benchmark methods in terms of computation time, total travel distance, number of local interactions between robots, and exploration rate across various degrees of communication dropouts. A scalability study in 3D environments showed a decrease in exploration time with MADE-Net with increasing team and environment sizes. The experiments presented highlight the effectiveness and robustness of our method.