arXiv subjects

Xuehui Wang

Publications and source records attributed to Xuehui Wang.

At least 19 recordsLinked to original sources

arXiv2026

Combating Textual Noise and Redundancy: Entropy-Aware Dense Visual Token Pruning

Visual token pruning is a crucial strategy for accelerating VLMs by compressing redundant image patches, yet existing methods often fail to preserve critical cues under dense instructions and fine-grained queries. In this paper, we investigate this failure and identify two underlying bottlenecks: the widespread dispersion of textual noise that corrupts dense cross-modal scoring, and the feature fragmentation inherent to standard token selection. To address these issues, we propose Entropy-Aware Dense Pruning (EADP), a framework that reformulates pruning as a structured compression problem. EADP first leverages statistical entropy to quantify and filter out textual noise, yielding a robust, fine-grained instruction relevance score. Subsequently, instead of naive Top-K selection, EADP casts token selection as a submodular maximization problem with a spatial prior, explicitly ensuring a holistic and non-redundant visual representation. Extensive experiments demonstrate that EADP improves the accuracy-efficiency trade-off of VLMs, robustly preserving fine-grained visual cues under strict token budgets while achieving SoTA performance on challenging multimodal benchmarks.

cs.CV↗

arXiv2026

Moment-Video: Diagnosing Temporal Fidelity of Video MLLMs on Momentary Visual Events

Video multimodal large language models (MLLMs) have made rapid progress on general and long-form video understanding, yet their ability to preserve brief answer-critical visual evidence remains underexplored. Many practical questions are determined by momentary visual events: localized actions or state transitions that may last only a few frames. Such evidence can be skipped by sparse frame sampling, suppressed by visual-token compression, or diluted by coarse temporal aggregation, causing failures that language-side reasoning cannot reliably recover. We introduce Moment-Video, a benchmark for diagnosing the temporal fidelity of video MLLMs through momentary visual event understanding. Each question is grounded in a localized, visually observable, and sampling-sensitive event, requiring models to notice, count, describe, or reason about transient evidence rather than rely on persistent objects, global scene context, or language priors. Moment-Video contains 1,000 human-verified video-QA pairs across 7 domains and 25 fine-grained subcategories, covering four task types: Temporal Occurrence, Temporal Counting, Action Description, and Temporal Reasoning. We evaluate 33 proprietary and open-source MLLMs on Moment-Video. The best-performing model, Seed-2.0-Pro, achieves only 39.6% overall accuracy, while most open-source models remain below 25%, revealing a substantial gap in momentary visual event understanding. Diagnostic analyses show that denser frame sampling improves some models but does not eliminate the bottleneck, and longer videos introduce stronger temporal-localization challenges. These findings suggest that current video MLLMs still lack temporally faithful representations for capturing, preserving, and using brief but decisive visual evidence.

cs.CV↗

arXiv2025

ScaleCUA: Scaling Open-Source Computer Use Agents with Cross-Platform Data

Vision-Language Models (VLMs) have enabled computer use agents (CUAs) that operate GUIs autonomously, showing great potential, yet progress is limited by the lack of large-scale, open-source computer use data and foundation models. In this work, we introduce ScaleCUA, a step toward scaling open-source CUAs. It offers a large-scale dataset spanning 6 operating systems and 3 task domains, built via a closed-loop pipeline uniting automated agents with human experts. Trained on this scaled-up data, ScaleCUA can operate seamlessly across platforms. Specifically, it delivers strong gains over baselines (+26.6 on WebArena-Lite-v2, +10.7 on ScreenSpot-Pro) and sets new state-of-the-art results (94.4% on MMBench-GUI L1-Hard, 60.6% on OSWorld-G, 47.4% on WebArena-Lite-v2). These findings underscore the power of data-driven scaling for general-purpose computer use agents. We will release data, models, and code to advance future research: https://github.com/OpenGVLab/ScaleCUA.

cs.CV↗

arXiv2025

InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency

We introduce InternVL 3.5, a new family of open-source multimodal models that significantly advances versatility, reasoning capability, and inference efficiency along the InternVL series. A key innovation is the Cascade Reinforcement Learning (Cascade RL) framework, which enhances reasoning through a two-stage process: offline RL for stable convergence and online RL for refined alignment. This coarse-to-fine training strategy leads to substantial improvements on downstream reasoning tasks, e.g., MMMU and MathVista. To optimize efficiency, we propose a Visual Resolution Router (ViR) that dynamically adjusts the resolution of visual tokens without compromising performance. Coupled with ViR, our Decoupled Vision-Language Deployment (DvD) strategy separates the vision encoder and language model across different GPUs, effectively balancing computational load. These contributions collectively enable InternVL3.5 to achieve up to a +16.0\% gain in overall reasoning performance and a 4.05$\times$ inference speedup compared to its predecessor, i.e., InternVL3. In addition, InternVL3.5 supports novel capabilities such as GUI interaction and embodied agency. Notably, our largest model, i.e., InternVL3.5-241B-A28B, attains state-of-the-art results among open-source MLLMs across general multimodal, reasoning, text, and agentic tasks -- narrowing the performance gap with leading commercial models like GPT-5. All models and code are publicly released.

cs.CV↗

arXiv2025

MMBench-GUI: Hierarchical Multi-Platform Evaluation Framework for GUI Agents

We introduce MMBench-GUI, a hierarchical benchmark for evaluating GUI automation agents across Windows, macOS, Linux, iOS, Android, and Web platforms. It comprises four levels: GUI Content Understanding, Element Grounding, Task Automation, and Task Collaboration, covering essential skills for GUI agents. In addition, we propose a novel Efficiency-Quality Area (EQA) metric to assess GUI agent execution efficiency in online automation scenarios. Through MMBench-GUI, we identify accurate visual grounding as a critical determinant of overall task success, emphasizing the substantial benefits of modular frameworks that integrate specialized grounding modules. Furthermore, to achieve reliable GUI automation, an agent requires strong task planning and cross-platform generalization abilities, with long-context memory, a broad action space, and long-term reasoning playing a critical role. More important, task efficiency remains a critically underexplored dimension, and all models suffer from substantial inefficiencies, with excessive redundant steps even when tasks are ultimately completed. The integration of precise localization, effective planning, and early stopping strategies is indispensable to enable truly efficient and scalable GUI automation. Our benchmark code, evaluation data, and running environment will be publicly available at https://github.com/open-compass/MMBench-GUI.

cs.CV↗

arXiv2025

Quasi-Periodic Optical Key-Enabled Hybrid Cryptography: Merging Diffractive Physics and Deep Learning for High-Dimensional Security

Optical encryption inherently provides strong security advantages, with hybrid optoelectronic systems offering additional degrees of freedom by integrating optical and algorithmic domains. However, existing optical encryption schemes heavily rely on electronic computation, limiting overall efficiency, while the physical keys are susceptible to damage, compromising both security and system stability. To overcome these challenges, we introduce the Quasi Periodic Optical Key (QPOK), which combines long range order with short range disorder, enabling enhanced security and robustness against damage within a single platform. By leveraging diffraction symmetry, our design enables optics-driven encryption, effectively shifting the optoelectronic balance toward photonic processing. Moreover, we innovatively apply deep learning to reconstruct the complex optical ciphertext field using only amplitude data and cryptographic keys, simultaneously achieving data compression and improved security. Within this framework, the key space includes continuously tunable parameters such as wavelength, propagation distance, phase modulation, and Q-POK geometry, significantly expanding cryptographic diversity. Our system also demonstrates robust cryptographic reliability by reducing inter-class distances by over 50% and tolerating up to 20% ciphertext loss. Our framework represents a new generation of physically grounded, algorithmically enhanced optical cryptosystems, laying a foundational pathway for scalable, hardware-integrated information security paradigms.

physics.optics↗

arXiv2025

ZeroGUI: Automating Online GUI Learning at Zero Human Cost

The rapid advancement of large Vision-Language Models (VLMs) has propelled the development of pure-vision-based GUI Agents, capable of perceiving and operating Graphical User Interfaces (GUI) to autonomously fulfill user instructions. However, existing approaches usually adopt an offline learning framework, which faces two core limitations: (1) heavy reliance on high-quality manual annotations for element grounding and action supervision, and (2) limited adaptability to dynamic and interactive environments. To address these limitations, we propose ZeroGUI, a scalable, online learning framework for automating GUI Agent training at Zero human cost. Specifically, ZeroGUI integrates (i) VLM-based automatic task generation to produce diverse training goals from the current environment state, (ii) VLM-based automatic reward estimation to assess task success without hand-crafted evaluation functions, and (iii) two-stage online reinforcement learning to continuously interact with and learn from GUI environments. Experiments on two advanced GUI Agents (UI-TARS and Aguvis) demonstrate that ZeroGUI significantly boosts performance across OSWorld and AndroidLab environments. The code is available at https://github.com/OpenGVLab/ZeroGUI.

cs.AI↗

arXiv2025

InstructSAM: A Training-Free Framework for Instruction-Oriented Remote Sensing Object Recognition

Language-Guided object recognition in remote sensing imagery is crucial for large-scale mapping and automated data annotation. However, existing open-vocabulary and visual grounding methods rely on explicit category cues, limiting their ability to handle complex or implicit queries that require advanced reasoning. To address this issue, we introduce a new suite of tasks, including Instruction-Oriented Object Counting, Detection, and Segmentation (InstructCDS), covering open-vocabulary, open-ended, and open-subclass scenarios. We further present EarthInstruct, the first InstructCDS benchmark for earth observation. It is constructed from two diverse remote sensing datasets with varying spatial resolutions and annotation rules across 20 categories, necessitating models to interpret dataset-specific instructions. Given the scarcity of semantically rich labeled data in remote sensing, we propose InstructSAM, a training-free framework for instruction-driven object recognition. InstructSAM leverages large vision-language models to interpret user instructions and estimate object counts, employs SAM2 for mask proposal, and formulates mask-label assignment as a binary integer programming problem. By integrating semantic similarity with counting constraints, InstructSAM efficiently assigns categories to predicted masks without relying on confidence thresholds. Experiments demonstrate that InstructSAM matches or surpasses specialized baselines across multiple tasks while maintaining near-constant inference time regardless of object count, reducing output tokens by 89% and overall runtime by over 32% compared to direct generation approaches. We believe the contributions of the proposed tasks, benchmark, and effective approach will advance future research in developing versatile object recognition systems.

cs.CV↗

arXiv2025

InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models

We introduce InternVL3, a significant advancement in the InternVL series featuring a native multimodal pre-training paradigm. Rather than adapting a text-only large language model (LLM) into a multimodal large language model (MLLM) that supports visual inputs, InternVL3 jointly acquires multimodal and linguistic capabilities from both diverse multimodal data and pure-text corpora during a single pre-training stage. This unified training paradigm effectively addresses the complexities and alignment challenges commonly encountered in conventional post-hoc training pipelines for MLLMs. To further improve performance and scalability, InternVL3 incorporates variable visual position encoding (V2PE) to support extended multimodal contexts, employs advanced post-training techniques such as supervised fine-tuning (SFT) and mixed preference optimization (MPO), and adopts test-time scaling strategies alongside an optimized training infrastructure. Extensive empirical evaluations demonstrate that InternVL3 delivers superior performance across a wide range of multi-modal tasks. In particular, InternVL3-78B achieves a score of 72.2 on the MMMU benchmark, setting a new state-of-the-art among open-source MLLMs. Its capabilities remain highly competitive with leading proprietary models, including ChatGPT-4o, Claude 3.5 Sonnet, and Gemini 2.5 Pro, while also maintaining strong pure-language proficiency. In pursuit of open-science principles, we will publicly release both the training data and model weights to foster further research and development in next-generation MLLMs.

cs.CV↗

arXiv2025

Secure Directional Modulation with Movable Antenna Array Aided by RIS

In this paper, to fully exploit the performance gains from moveable antennas (MAs) and reconfigurable intelligent surface (RIS), a RIS-aided directional modulation \textcolor{blue}{(DM)} network with movable antenna at base station (BS) is established Based on the principle of DM, a BS equipped with MAs transmits legitimate information to a single-antenna user (Bob) while exploiting artificial noise (AN) to degrade signal reception at the eavesdropper (Eve). The combination of AN and transmission beamforming vectors is modeled as joint beamforming vector (JBV) to achieve optimal power allocation. The objective is to maximize the achievable secrecy rate (SR) by optimizing MAs antenna position, phase shift matrix (PSM) of RIS, and JBV. The limited movable range (MR) and discrete candidate positions of the MAs at the BS are considered, which renders the optimization problem non-convex. To address these challenges, an optimization method under perfect channel state information (CSI) is firstly designed, in which the MAs antenna positions are obtained using compressive sensing (CS) technology, and JBV and PSM are iteratively optimized. Then, the design method and SR performance under imperfect CSI is investigated. The proposed algorithms have fewer iterations and lower complexity. Simulation results demonstrate that MAs outperform fixed-position antennas in SR performance when there is an adequately large MR available.

eess.SP↗

arXiv2025

Generalized Tensor-based Parameter-Efficient Fine-Tuning via Lie Group Transformations

Adapting pre-trained foundation models for diverse downstream tasks is a core practice in artificial intelligence. However, the wide range of tasks and high computational costs make full fine-tuning impractical. To overcome this, parameter-efficient fine-tuning (PEFT) methods like LoRA have emerged and are becoming a growing research focus. Despite the success of these methods, they are primarily designed for linear layers, focusing on two-dimensional matrices while largely ignoring higher-dimensional parameter spaces like convolutional kernels. Moreover, directly applying these methods to higher-dimensional parameter spaces often disrupts their structural relationships. Given the rapid advancements in matrix-based PEFT methods, rather than designing a specialized strategy, we propose a generalization that extends matrix-based PEFT methods to higher-dimensional parameter spaces without compromising their structural properties. Specifically, we treat parameters as elements of a Lie group, with updates modeled as perturbations in the corresponding Lie algebra. These perturbations are mapped back to the Lie group through the exponential map, ensuring smooth, consistent updates that preserve the inherent structure of the parameter space. Extensive experiments on computer vision and natural language processing validate the effectiveness and versatility of our approach, demonstrating clear improvements over existing methods.

cs.LG↗

arXiv2024

ICM-Assistant: Instruction-tuning Multimodal Large Language Models for Rule-based Explainable Image Content Moderation

Controversial contents largely inundate the Internet, infringing various cultural norms and child protection standards. Traditional Image Content Moderation (ICM) models fall short in producing precise moderation decisions for diverse standards, while recent multimodal large language models (MLLMs), when adopted to general rule-based ICM, often produce classification and explanation results that are inconsistent with human moderators. Aiming at flexible, explainable, and accurate ICM, we design a novel rule-based dataset generation pipeline, decomposing concise human-defined rules and leveraging well-designed multi-stage prompts to enrich short explicit image annotations. Our ICM-Instruct dataset includes detailed moderation explanation and moderation Q-A pairs. Built upon it, we create our ICM-Assistant model in the framework of rule-based ICM, making it readily applicable in real practice. Our ICM-Assistant model demonstrates exceptional performance and flexibility. Specifically, it significantly outperforms existing approaches on various sources, improving both the moderation classification (36.8% on average) and moderation explanation quality (26.6% on average) consistently over existing MLLMs. Code/Data is available at https://github.com/zhaoyuzhi/ICM-Assistant.

cs.CV↗

arXiv2024

Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling

We introduce InternVL 2.5, an advanced multimodal large language model (MLLM) series that builds upon InternVL 2.0, maintaining its core model architecture while introducing significant enhancements in training and testing strategies as well as data quality. In this work, we delve into the relationship between model scaling and performance, systematically exploring the performance trends in vision encoders, language models, dataset sizes, and test-time configurations. Through extensive evaluations on a wide range of benchmarks, including multi-discipline reasoning, document understanding, multi-image / video understanding, real-world comprehension, multimodal hallucination detection, visual grounding, multilingual capabilities, and pure language processing, InternVL 2.5 exhibits competitive performance, rivaling leading commercial models such as GPT-4o and Claude-3.5-Sonnet. Notably, our model is the first open-source MLLMs to surpass 70% on the MMMU benchmark, achieving a 3.7-point improvement through Chain-of-Thought (CoT) reasoning and showcasing strong potential for test-time scaling. We hope this model contributes to the open-source community by setting new standards for developing and applying multimodal AI systems. HuggingFace demo see https://huggingface.co/spaces/OpenGVLab/InternVL

cs.CV↗

arXiv2024

Which Channel in 6G, Low-rank or Full-rank, more needs RIS from a Perspective of DoF?

Reconfigurable intelligent surface (RIS), as an efficient tool to improve receive signal-to-noise ratio, extend coverage and create more spatial diversity, is viewed as a most promising technique for the future wireless networks like 6G. As you know, RIS is very suitable for a special wireless scenario with wireless link between BS and users being completely blocked, i.e., no link. In this paper, we extend its applications to a general scenario, i.e., rank-deficient channel, particularly some extremely low-rank ones such as no link, and line-of-sight (LoS, rank-one). Actually, there are several potential important low-rank applications like low-altitude, satellite, UAV, marine, and deep-space communications. In such a situation, it is found that RIS may make a dramatic degrees of freedom (DoF) enhancement over no RIS. By using a distributed RISs placement, the DoF of channel from BS to user in LoS channel may be even boosted from a low-rank like 0/1 to full-rank. This will achieve an extremely rate improvement via spatial parallel multiple-stream transmission from BS to user. In this paper, we present a complete review of making an in-depth discussion on DoF effect of RIS.

eess.SP↗

arXiv2024

DoF Analysis and Beamforming Design for Active IRS-aided Multi-user MIMO Wireless Communication in Rank-deficient Channels

Due to its ability of significantly improving data rate, intelligent reflecting surface (IRS) will be a potential crucial technique for the future generation wireless networks like 6G. In this paper, we will focus on the analysis of degree of freedom (DoF) in IRS-aided multi-user MIMO network. Firstly, the DoF upper bound of IRS-aided single-user MIMO network, i.e., the achievable maximum DoF of such a system, is derived, and the corresponding results are extended to the case of IRS-aided multiuser MIMO by using the matrix rank inequalities. In particular, in serious rank-deficient, also called low-rank, channels like line-of-sight (LoS), the network DoF may doubles over no-IRS with the help of IRS. To verify the rate performance gain from augmented DoF, three closed-form beamforming methods, null-space projection plus maximize transmit power and maximize receive power (NSP-MTP-MRP), Schmidt orthogonalization plus minimum mean square error (SO-MMSE) and two-layer leakage plus MMSE (TLL-MMSE) are proposed to achieve the maximum DoF. Simulation results shows that IRS does make a dramatic rate enhancement. For example, in a serious deficient channel, the sum-rate of the proposed TLL-MMSE aided by IRS is about twice that of no IRS. This means that IRS may achieve a significant DoF improvement in such a channel.

eess.SP↗

arXiv2024

Joint Power Allocation and Beamforming Design for Active IRS-Aided Directional Modulation Secure Systems

Since the secrecy rate (SR) performance improvement obtained by secure directional modulation (DM) network is limited, an active intelligent reflective surface (IRS)-assisted DM network is considered to attain a high SR. To address the SR maximization problem, a novel method based on Lagrangian dual transform and closed-form fractional programming algorithm (LDT-CFFP) is proposed, where the solutions to base station (BS) beamforming vectors and IRS reflection coefficient matrix are achieved. However, the computational complexity of LDT-CFFP method is high . To reduce its complexity, a blocked IRS-assisted DM network is designed. To meet the requirements of the network performance, a power allocation (PA) strategy is proposed and adopted in the system. Specifically, the system power between BS and IRS, as well as the transmission power for confidential messages (CM) and artificial noise (AN) from the BS, are allocated separately. Then we put forward null-space projection (NSP) method, maximum-ratio-reflecting (MRR) algorithm and PA strategy (NSP-MRR-PA) to solve the SR maximization problem. The CF solutions to BS beamforming vectors and IRS reflection coefficient matrix are respectively attained via NSP and MRR algorithms. For the PA factors, we take advantage of exhaustive search (ES) algorithm, particle swarm optimization (PSO) and simulated annealing (SA) algorithm to search for the solutions. From simulation results, it is verified that the LDT-CFFP method derives a higher SR gain over NSP-MRR-PA method. For NSP-MRR-PA method, the number of IRS units in each block possesses a significant SR performance. In addition, the application PA strategies, namely ES, PSO, SA methods outperforms the other PA strategies with fixed PA factors.

eess.SP↗

arXiv2024

Maintaining Structural Integrity in Parameter Spaces for Parameter Efficient Fine-tuning

Adapting pre-trained foundation models for various downstream tasks has been prevalent in artificial intelligence. Due to the vast number of tasks and high costs, adjusting all parameters becomes unfeasible. To mitigate this, several fine-tuning techniques have been developed to update the pre-trained model weights in a more resource-efficient manner, such as through low-rank adjustments. Yet, almost all of these methods focus on linear weights, neglecting the intricacies of parameter spaces in higher dimensions like 4D. Alternatively, some methods can be adapted for high-dimensional parameter space by compressing changes in the original space into two dimensions and then employing low-rank matrix adaptations. However, these approaches destructs the structural integrity of the involved high-dimensional spaces. To tackle the diversity of dimensional spaces across different foundation models and provide a more precise representation of the changes within these spaces, this paper introduces a generalized parameter-efficient fine-tuning framework, designed for various dimensional parameter space. Specifically, our method asserts that changes in each dimensional parameter space are based on a low-rank core space which maintains the consistent topological structure with the original space. It then models the changes through this core space alongside corresponding weights to reconstruct alterations in the original space. It effectively preserves the structural integrity of the change of original N-dimensional parameter space, meanwhile models it via low-rank tensor adaptation. Extensive experiments on computer vision, natural language processing and multi-modal tasks validate the effectiveness of our method.

cs.CV↗

arXiv2024

Multi-stream Transmission for Directional Modulation Network via Distributed Multi-UAV-aided Multi-active-IRS

Active intelligent reflecting surface (IRS) is a revolutionary technique for the future 6G networks. The conventional far-field single-IRS-aided directional modulation(DM) networks have only one (no direct path) or two (existing direct path) degrees of freedom (DoFs). This means that there are only one or two streams transmitted simultaneously from base station to user and will seriously limit its rate gain achieved by IRS. How to create multiple DoFs more than two for DM? In this paper, single large-scale IRS is divided to multiple small IRSs and a novel multi-IRS-aided multi-stream DM network is proposed to achieve a point-to-point multi-stream transmission by creating $K$ ($\geq3$) DoFs, where multiple small IRSs are placed distributively via multiple unmanned aerial vehicles (UAVs). The null-space projection, zero-forcing (ZF) and phase alignment are adopted to design the transmit beamforming vector, receive beamforming vector and phase shift matrix (PSM), respectively, called NSP-ZF-PA. Here, $K$ PSMs and their corresponding beamforming vectors are independently optimized. The weighted minimum mean-square error (WMMSE) algorithm is involved in alternating iteration for the optimization variables by introducing the power constraint on IRS, named WMMSE-PC, where the majorization-minimization (MM) algorithm is used to solve the total PSM. To achieve a lower computational complexity, a maximum trace method, called Max-TR-SVD, is proposed by optimize the PSM of all IRSs. Numerical simulation results has shown that the proposed NSP-ZF-PA performs much better than Max-TR-SVD in terms of rate. In particular, the rate of NSP-ZF-PA with sixteen small IRSs is about five times that of NSP-ZF-PA with combining all small IRSs as a single large IRS. Thus, a dramatic rate enhancement may be achieved by multiple distributed IRSs.

eess.SP↗