arXiv subjects

Xiaohan Zhang

Publications and source records attributed to Xiaohan Zhang.

At least 73 records · Page 4Linked to original sources

arXiv2025

Melodia: Training-Free Music Editing Guided by Attention Probing in Diffusion Models

Text-to-music generation technology is progressing rapidly, creating new opportunities for musical composition and editing. However, existing music editing methods often fail to preserve the source music's temporal structure, including melody and rhythm, when altering particular attributes like instrument, genre, and mood. To address this challenge, this paper conducts an in-depth probing analysis on attention maps within AudioLDM 2, a diffusion-based model commonly used as the backbone for existing music editing methods. We reveal a key finding: cross-attention maps encompass details regarding distinct musical characteristics, and interventions on these maps frequently result in ineffective modifications. In contrast, self-attention maps are essential for preserving the temporal structure of the source music during its conversion into the target music. Building upon this understanding, we present Melodia, a training-free technique that selectively manipulates self-attention maps in particular layers during the denoising process and leverages an attention repository to store source music information, achieving accurate modification of musical characteristics while preserving the original structure without requiring textual descriptions of the source music. Additionally, we propose two novel metrics to better evaluate music editing methods. Both objective and subjective experiments demonstrate that our approach achieves superior results in terms of textual adherence and structural integrity across various datasets. This research enhances comprehension of internal mechanisms within music generation models and provides improved control for music creation.

cs.SD↗

arXiv2025

SceneJailEval: A Scenario-Adaptive Multi-Dimensional Framework for Jailbreak Evaluation

Accurate jailbreak evaluation is critical for LLM red team testing and jailbreak research. Mainstream methods rely on binary classification (string matching, toxic text classifiers, and LLM-based methods), outputting only "yes/no" labels without quantifying harm severity. Emerged multi-dimensional frameworks (e.g., Security Violation, Relative Truthfulness and Informativeness) use unified evaluation standards across scenarios, leading to scenario-specific mismatches (e.g., "Relative Truthfulness" is irrelevant to "hate speech"), undermining evaluation accuracy. To address these, we propose SceneJailEval, with key contributions: (1) A pioneering scenario-adaptive multi-dimensional framework for jailbreak evaluation, overcoming the critical "one-size-fits-all" limitation of existing multi-dimensional methods, and boasting robust extensibility to seamlessly adapt to customized or emerging scenarios. (2) A novel 14-scenario dataset featuring rich jailbreak variants and regional cases, addressing the long-standing gap in high-quality, comprehensive benchmarks for scenario-adaptive evaluation. (3) SceneJailEval delivers state-of-the-art performance with an F1 score of 0.917 on our full-scenario dataset (+6% over SOTA) and 0.995 on JBB (+3% over SOTA), breaking through the accuracy bottleneck of existing evaluation methods in heterogeneous scenarios and solidifying its superiority.

cs.CL↗

arXiv2025

Depth-Consistent 3D Gaussian Splatting via Physical Defocus Modeling and Multi-View Geometric Supervision

Three-dimensional reconstruction in scenes with extreme depth variations remains challenging due to inconsistent supervisory signals between near-field and far-field regions. Existing methods fail to simultaneously address inaccurate depth estimation in distant areas and structural degradation in close-range regions. This paper proposes a novel computational framework that integrates depth-of-field supervision and multi-view consistency supervision to advance 3D Gaussian Splatting. Our approach comprises two core components: (1) Depth-of-field Supervision employs a scale-recovered monocular depth estimator (e.g., Metric3D) to generate depth priors, leverages defocus convolution to synthesize physically accurate defocused images, and enforces geometric consistency through a novel depth-of-field loss, thereby enhancing depth fidelity in both far-field and near-field regions; (2) Multi-View Consistency Supervision employing LoFTR-based semi-dense feature matching to minimize cross-view geometric errors and enforce depth consistency via least squares optimization of reliable matched points. By unifying defocus physics with multi-view geometric constraints, our method achieves superior depth fidelity, demonstrating a 0.8 dB PSNR improvement over the state-of-the-art method on the Waymo Open Dataset. This framework bridges physical imaging principles and learning-based depth regularization, offering a scalable solution for complex depth stratification in urban environments.

cs.CV↗

arXiv2025

LLM-GROP: Visually Grounded Robot Task and Motion Planning with Large Language Models

Task planning and motion planning are two of the most important problems in robotics, where task planning methods help robots achieve high-level goals and motion planning methods maintain low-level feasibility. Task and motion planning (TAMP) methods interleave the two processes of task planning and motion planning to ensure goal achievement and motion feasibility. Within the TAMP context, we are concerned with the mobile manipulation (MoMa) of multiple objects, where it is necessary to interleave actions for navigation and manipulation. In particular, we aim to compute where and how each object should be placed given underspecified goals, such as ``set up dinner table with a fork, knife and plate.'' We leverage the rich common sense knowledge from large language models (LLMs), e.g., about how tableware is organized, to facilitate both task-level and motion-level planning. In addition, we use computer vision methods to learn a strategy for selecting base positions to facilitate MoMa behaviors, where the base position corresponds to the robot's ``footprint'' and orientation in its operating space. Altogether, this article provides a principled TAMP framework for MoMa tasks that accounts for common sense about object rearrangement and is adaptive to novel situations that include many objects that need to be moved. We performed quantitative experiments in both real-world settings and simulated environments. We evaluated the success rate and efficiency in completing long-horizon object rearrangement tasks. While the robot completed 84.4\% real-world object rearrangement trials, subjective human evaluations indicated that the robot's performance is still lower than experienced human waiters.

cs.RO↗

arXiv2025

RaGS: Unleashing 3D Gaussian Splatting from 4D Radar and Monocular Cues for 3D Object Detection

4D millimeter-wave radar is a promising sensing modality for autonomous driving, yet effective 3D object detection from 4D radar and monocular images remains challenging. Existing fusion approaches either rely on instance proposals lacking global context or dense BEV grids constrained by rigid structures, lacking a flexible and adaptive representation for diverse scenes. To address this, we propose RaGS, the first framework that leverages 3D Gaussian Splatting (GS) to fuse 4D radar and monocular cues for 3D object detection. 3D GS models the scene as a continuous field of Gaussians, enabling dynamic resource allocation to foreground objects while maintaining flexibility and efficiency. Moreover, the velocity dimension of 4D radar provides motion cues that help anchor and refine the spatial distribution of Gaussians. Specifically, RaGS adopts a cascaded pipeline to construct and progressively refine the Gaussian field. It begins with Frustum-based Localization Initiation (FLI), which unprojects foreground pixels to initialize coarse Gaussian centers. Then, Iterative Multimodal Aggregation (IMA) explicitly exploits image semantics and implicitly integrates 4D radar velocity geometry to refine the Gaussians within regions of interest. Finally, Multi-level Gaussian Fusion (MGF) renders the Gaussian field into hierarchical BEV features for 3D object detection. By dynamically focusing on sparse and informative regions, RaGS achieves object-centric precision and comprehensive scene perception. Extensive experiments on View-of-Delft, TJ4DRadSet, and OmniHD-Scenes demonstrate its robustness and SOTA performance. Code will be released.

cs.CV↗

arXiv2025

Data-Efficient Multitask DAgger

Generalist robot policies that can perform many tasks typically require extensive expert data or simulations for training. In this work, we propose a novel Data-Efficient multitask DAgger framework that distills a single multitask policy from multiple task-specific expert policies. Our approach significantly increases the overall task success rate by actively focusing on tasks where the multitask policy underperforms. The core of our method is a performance-aware scheduling strategy that tracks how much each task's learning process benefits from the amount of data, using a Kalman filter-based estimator to robustly decide how to allocate additional demonstrations across tasks. We validate our approach on MetaWorld, as well as a suite of diverse drawer-opening tasks in IsaacLab. The resulting policy attains high performance across all tasks while using substantially fewer expert demonstrations, and the visual policy learned with our method in simulation shows better performance than naive DAgger and Behavior Cloning when transferring zero-shot to a real robot without using real data.

cs.LG↗

arXiv2025

ParticleFormer: A 3D Point Cloud World Model for Multi-Object, Multi-Material Robotic Manipulation

3D world models (i.e., learning-based 3D dynamics models) offer a promising approach to generalizable robotic manipulation by capturing the underlying physics of environment evolution conditioned on robot actions. However, existing 3D world models are primarily limited to single-material dynamics using a particle-based Graph Neural Network model, and often require time-consuming 3D scene reconstruction to obtain 3D particle tracks for training. In this work, we present ParticleFormer, a Transformer-based point cloud world model trained with a hybrid point cloud reconstruction loss, supervising both global and local dynamics features in multi-material, multi-object robot interactions. ParticleFormer captures fine-grained multi-object interactions between rigid, deformable, and flexible materials, trained directly from real-world robot perception data without an elaborate scene reconstruction. We demonstrate the model's effectiveness both in 3D scene forecasting tasks, and in downstream manipulation tasks using a Model Predictive Control (MPC) policy. In addition, we extend existing dynamics learning benchmarks to include diverse multi-material, multi-object interaction scenarios. We validate our method on six simulation and three real-world experiments, where it consistently outperforms leading baselines by achieving superior dynamics prediction accuracy and less rollout error in downstream visuomotor tasks. Experimental videos are available at https://suninghuang19.github.io/particleformer_page/.

cs.RO↗

arXiv2025

LVBench: An Extreme Long Video Understanding Benchmark

Recent progress in multimodal large language models has markedly enhanced the understanding of short videos (typically under one minute), and several evaluation datasets have emerged accordingly. However, these advancements fall short of meeting the demands of real-world applications such as embodied intelligence for long-term decision-making, in-depth movie reviews and discussions, and live sports commentary, all of which require comprehension of long videos spanning several hours. To address this gap, we introduce LVBench, a benchmark specifically designed for long video understanding. Our dataset comprises publicly sourced videos and encompasses a diverse set of tasks aimed at long video comprehension and information extraction. LVBench is designed to challenge multimodal models to demonstrate long-term memory and extended comprehension capabilities. Our extensive evaluations reveal that current multimodal models still underperform on these demanding long video understanding tasks. Through LVBench, we aim to spur the development of more advanced models capable of tackling the complexities of long video comprehension. Our data and code are publicly available at: https://lvbench.github.io.

cs.CV↗

arXiv2025

MASteer: Multi-Agent Adaptive Steer Strategy for End-to-End LLM Trustworthiness Repair

Large Language Models (LLMs) face persistent and evolving trustworthiness issues, motivating developers to seek automated and flexible repair methods that enable convenient deployment across diverse scenarios. Existing repair methods like supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) are costly and slow, while prompt engineering lacks robustness and scalability. Representation engineering, which steers model behavior by injecting targeted concept vectors during inference, offers a lightweight, training-free alternative. However, current approaches depend on manually crafted samples and fixed steering strategies, limiting automation and adaptability. To overcome these challenges, we propose MASteer, the first end-to-end framework for trustworthiness repair in LLMs based on representation engineering. MASteer integrates two core components: AutoTester, a multi-agent system that generates diverse, high-quality steer samples tailored to developer needs; and AutoRepairer, which constructs adaptive steering strategies with anchor vectors for automated, context-aware strategy selection during inference. Experiments on standard and customized trustworthiness tasks show MASteer consistently outperforms baselines, improving metrics by 15.36% on LLaMA-3.1-8B-Chat and 4.21% on Qwen-3-8B-Chat, while maintaining general model capabilities. MASteer demonstrates strong robustness, generalization, and practical value for scalable, efficient trustworthiness repair.

cs.AI↗

arXiv2025

GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models

We present GLM-4.5, an open-source Mixture-of-Experts (MoE) large language model with 355B total parameters and 32B activated parameters, featuring a hybrid reasoning method that supports both thinking and direct response modes. Through multi-stage training on 23T tokens and comprehensive post-training with expert model iteration and reinforcement learning, GLM-4.5 achieves strong performance across agentic, reasoning, and coding (ARC) tasks, scoring 70.1% on TAU-Bench, 91.0% on AIME 24, and 64.2% on SWE-bench Verified. With much fewer parameters than several competitors, GLM-4.5 ranks 3rd overall among all evaluated models and 2nd on agentic benchmarks. We release both GLM-4.5 (355B parameters) and a compact version, GLM-4.5-Air (106B parameters), to advance research in reasoning and agentic AI systems. Code, models, and more information are available at https://github.com/zai-org/GLM-4.5.

cs.CL↗

arXiv2025

VICI: VLM-Instructed Cross-view Image-localisation

In this paper, we present a high-performing solution to the UAVM 2025 Challenge, which focuses on matching narrow FOV street-level images to corresponding satellite imagery using the University-1652 dataset. As panoramic Cross-View Geo-Localisation nears peak performance, it becomes increasingly important to explore more practical problem formulations. Real-world scenarios rarely offer panoramic street-level queries; instead, queries typically consist of limited-FOV images captured with unknown camera parameters. Our work prioritises discovering the highest achievable performance under these constraints, pushing the limits of existing architectures. Our method begins by retrieving candidate satellite image embeddings for a given query, followed by a re-ranking stage that selectively enhances retrieval accuracy within the top candidates. This two-stage approach enables more precise matching, even under the significant viewpoint and scale variations inherent in the task. Through experimentation, we demonstrate that our approach achieves competitive results -specifically attaining R@1 and R@10 retrieval rates of \topone\% and \topten\% respectively. This underscores the potential of optimised retrieval and re-ranking strategies in advancing practical geo-localisation performance. Code is available at https://github.com/tavisshore/VICI.

cs.CV↗

arXiv2025

TraGraph-GS: Trajectory Graph-based Gaussian Splatting for Arbitrary Large-Scale Scene Rendering

High-quality novel view synthesis for large-scale scenes presents a challenging dilemma in 3D computer vision. Existing methods typically partition large scenes into multiple regions, reconstruct a 3D representation using Gaussian splatting for each region, and eventually merge them for novel view rendering. They can accurately render specific scenes, yet they do not generalize effectively for two reasons: (1) rigid spatial partition techniques struggle with arbitrary camera trajectories, and (2) the merging of regions results in Gaussian overlap to distort texture details. To address these challenges, we propose TraGraph-GS, leveraging a trajectory graph to enable high-precision rendering for arbitrarily large-scale scenes. We present a spatial partitioning method for large-scale scenes based on graphs, which incorporates a regularization constraint to enhance the rendering of textures and distant objects, as well as a progressive rendering strategy to mitigate artifacts caused by Gaussian overlap. Experimental results demonstrate its superior performance both on four aerial and four ground datasets and highlight its remarkable efficiency: our method achieves an average improvement of 1.86 dB in PSNR on aerial datasets and 1.62 dB on ground datasets compared to state-of-the-art approaches.

cs.CV↗

arXiv2025

AlignMMBench: Evaluating Chinese Multimodal Alignment in Large Vision-Language Models

Evaluating the alignment capabilities of large Vision-Language Models (VLMs) is essential for determining their effectiveness as helpful assistants. However, existing benchmarks primarily focus on basic abilities using nonverbal methods, such as yes-no and multiple-choice questions. In this paper, we address this gap by introducing AlignMMBench, which provides more nuanced evaluations of alignment capabilities and is the first benchmark specifically designed for Chinese visual contexts. This benchmark is meticulously curated from real-world scenarios and internet sources, encompassing thirteen specific tasks across three categories, and includes both single-turn and multi-turn dialogue scenarios. Incorporating a prompt rewrite strategy, AlignMMBench encompasses 1,054 images and 4,978 question-answer pairs. To facilitate the evaluation pipeline, we develop CritiqueVLM, a rule-calibrated evaluator that exceeds GPT-4's evaluation ability. Additionally, we measure the "alignment score", a quantitative metric designed to assess the robustness and stability of models across diverse prompts. Finally, we evaluate the performance of representative VLMs on AlignMMBench, offering insights into the capabilities and limitations of different VLM architectures. The evaluation code and data are available at https://github.com/THUDM/AlignMMBench.

cs.CL↗

arXiv2025

CDFormer: Cross-Domain Few-Shot Object Detection Transformer Against Feature Confusion

Cross-domain few-shot object detection (CD-FSOD) aims to detect novel objects across different domains with limited class instances. Feature confusion, including object-background confusion and object-object confusion, presents significant challenges in both cross-domain and few-shot settings. In this work, we introduce CDFormer, a cross-domain few-shot object detection transformer against feature confusion, to address these challenges. The method specifically tackles feature confusion through two key modules: object-background distinguishing (OBD) and object-object distinguishing (OOD). The OBD module leverages a learnable background token to differentiate between objects and background, while the OOD module enhances the distinction between objects of different classes. Experimental results demonstrate that CDFormer outperforms previous state-of-the-art approaches, achieving 12.9% mAP, 11.0% mAP, and 10.4% mAP improvements under the 1/5/10 shot settings, respectively, when fine-tuned.

cs.CV↗

arXiv2025

Metamon-GS: Enhancing Representability with Variance-Guided Densification and Light Encoding

The introduction of 3D Gaussian Splatting (3DGS) has advanced novel view synthesis by utilizing Gaussians to represent scenes. Encoding Gaussian point features with anchor embeddings has significantly enhanced the performance of newer 3DGS variants. While significant advances have been made, it is still challenging to boost rendering performance. Feature embeddings have difficulty accurately representing colors from different perspectives under varying lighting conditions, which leads to a washed-out appearance. Another reason is the lack of a proper densification strategy that prevents Gaussian point growth in thinly initialized areas, resulting in blurriness and needle-shaped artifacts. To address them, we propose Metamon-GS, from innovative viewpoints of variance-guided densification strategy and multi-level hash grid. The densification strategy guided by variance specifically targets Gaussians with high gradient variance in pixels and compensates for the importance of regions with extra Gaussians to improve reconstruction. The latter studies implicit global lighting conditions and accurately interprets color from different perspectives and feature embeddings. Our thorough experiments on publicly available datasets show that Metamon-GS surpasses its baseline model and previous versions, delivering superior quality in rendering novel views.

cs.CV↗

arXiv2025

Safe Text-to-Image Generation: Simply Sanitize the Prompt Embedding

In recent years, text-to-image (T2I) generation models have made significant progress in generating high-quality images that align with text descriptions. However, these models also face the risk of unsafe generation, potentially producing harmful content that violates usage policies, such as explicit material. Existing safe generation methods typically focus on suppressing inappropriate content by erasing undesired concepts from visual representations, while neglecting to sanitize the textual representation. Although these methods help mitigate the risk of misuse to some extent, their robustness remains insufficient when dealing with adversarial attacks. Given that semantic consistency between input text and output image is a core requirement of T2I models, we identify that textual representations are likely the primary source of unsafe generation. To this end, we propose Embedding Sanitizer (ES), which enhances the safety of T2I models by sanitizing inappropriate concepts in prompt embeddings. To our knowledge, ES is the first interpretable safe generation framework that assigns a score to each token in the prompt to indicate its potential harmfulness. In addition, ES adopts a plug-and-play modular design, offering compatibility for seamless integration with various T2I models and other safeguards. Evaluations on five prompt benchmarks show that ES outperforms eleven existing safeguard baselines, achieving state-of-the-art robustness while maintaining high-quality image generation.

cs.CR↗

arXiv2025

One-Dimensional Potassium Chains on Silicon Nanoribbons

Silicon nanoribbons (SiNRs), characterized by a pentagonal structure composed of silicon atoms, host one-dimensional (1D) Dirac Fermions and serve as a minimalist atomic template for adsorbing various heteroatoms. Alkali-metal (AM) atoms, such as Na and K, with electronic structures comparable to those of hydrogen are of particular interest for such adsorption studies. However, the adsorption of AM atoms on SiNRs and its tunation on the properties of SiNRs have not yet been fully explored. In this study, we examined the adsorption of K atoms on high-aspect-ratio SiNRs and the resultant electronic properties using a combination of scanning tunneling microscopy (STM) and density functional theory calculations. K atoms prefer to adsorb on double- and multi-stranded SiNRs owing to the low adsorption energies at these sites. Each K atom and its three nearest Si atoms exhibit a triangular morphology resulting from charge transfer between K and Si atoms, as verified by theoretical calculations. As the K coverage of the SiNRs increased, the K atoms organize into 1D zigzag chains on the SiNRs. Moreover, K adsorption on the SiNRs was determined to be reversible. The deposition of K atoms on the SiNRs was achieved using a voltage pulse of the STM tip, without damaging the SiNRs structure. In addition, K adsorption effectively modulates the Dirac cone position of the SiNRs relative to the Fermi level. This study unveils the adsorption mechanism of AM atoms on SiNRs, providing a useful approach for heteroatom adsorption on other nanoribbons.

cond-mat.mes-hall↗

arXiv2025

CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer

We present CogVideoX, a large-scale text-to-video generation model based on diffusion transformer, which can generate 10-second continuous videos aligned with text prompt, with a frame rate of 16 fps and resolution of 768 * 1360 pixels. Previous video generation models often had limited movement and short durations, and is difficult to generate videos with coherent narratives based on text. We propose several designs to address these issues. First, we propose a 3D Variational Autoencoder (VAE) to compress videos along both spatial and temporal dimensions, to improve both compression rate and video fidelity. Second, to improve the text-video alignment, we propose an expert transformer with the expert adaptive LayerNorm to facilitate the deep fusion between the two modalities. Third, by employing a progressive training and multi-resolution frame pack technique, CogVideoX is adept at producing coherent, long-duration, different shape videos characterized by significant motions. In addition, we develop an effective text-video data processing pipeline that includes various data preprocessing strategies and a video captioning method, greatly contributing to the generation quality and semantic alignment. Results show that CogVideoX demonstrates state-of-the-art performance across both multiple machine metrics and human evaluations. The model weight of both 3D Causal VAE, Video caption model and CogVideoX are publicly available at https://github.com/THUDM/CogVideo.

cs.CV↗