AgentHallu: Benchmarking Automated Hallucination Attribution of LLM-based Agents

As LLM-based agents operate over sequential multi-step reasoning, hallucinations arising at intermediate steps risk propagating along the trajectory, thus degrading overall reliability. Unlike hallucination detection in single-turn responses, diagnosing hallucinations in multi-step workflows requires identifying which step causes the initial divergence. To fill this gap, we propose a new research task, automated hallucination attribution of LLM-based agents, aiming to identify the step responsible for the hallucination and explain why. To support this task, we introduce AgentHallu, a comprehensive benchmark with: (1) 693 high-quality trajectories spanning 7 agent frameworks and 5 domains, (2) a hallucination taxonomy organized into 5 categories (Planning, Retrieval, Reasoning, Human-Interaction, and Tool-Use) and 14 sub-categories, and (3) multi-level annotations curated by humans, covering binary labels, hallucination-responsible steps, and causal explanations. We evaluate 13 leading models, and results show the task is challenging even for top-tier models (like GPT-5, Gemini-2.5-Pro). The best-performing model achieves only 41.1\% step localization accuracy, where tool-use hallucinations are the most challenging at just 11.6\%. We believe AgentHallu will catalyze future research into developing robust, transparent, and reliable agentic systems.

EgoReAct: Egocentric Video-Driven 3D Human Reaction Generation

Humans exhibit adaptive, context-sensitive responses to egocentric visual input. However, faithfully modeling such reactions from egocentric video remains challenging due to the dual requirements of strictly causal generation and precise 3D spatial alignment. To tackle this problem, we first construct the Human Reaction Dataset (HRD) to address data scarcity and misalignment by building a spatially aligned egocentric video-reaction dataset, as existing datasets (e.g., ViMo) suffer from significant spatial inconsistency between the egocentric video and reaction motion, e.g., dynamically moving motions are always paired with fixed-camera videos. Leveraging HRD, we present EgoReAct, the first autoregressive framework that generates 3D-aligned human reaction motions from egocentric video streams in real-time. We first compress the reaction motion into a compact yet expressive latent space via a Vector Quantised-Variational AutoEncoder and then train a Generative Pre-trained Transformer for reaction generation from the visual input. EgoReAct incorporates 3D dynamic features, i.e., metric depth, and head dynamics during the generation, which effectively enhance spatial grounding. Extensive experiments demonstrate that EgoReAct achieves remarkably higher realism, spatial consistency, and generation efficiency compared with prior methods, while maintaining strict causality during generation. We will release code, models, and data upon acceptance.

FRIEDA: Benchmarking Multi-Step Cartographic Reasoning in Vision-Language Models

Cartographic reasoning is the skill of interpreting geographic relationships by aligning legends, map scales, compass directions, map texts, and geometries across one or more map images. Although essential as a concrete cognitive capability and for critical tasks such as disaster response and urban planning, it remains largely unevaluated. Building on progress in chart and infographic understanding, recent large vision language model studies on map visual question-answering often treat maps as a special case of charts. In contrast, map VQA demands comprehension of layered symbology (e.g., symbols, geometries, and text labels) as well as spatial relations tied to orientation and distance that often span multiple maps and are not captured by chart-style evaluations. To address this gap, we introduce FRIEDA, a benchmark for testing complex open-ended cartographic reasoning in LVLMs. FRIEDA sources real map images from documents and reports in various domains and geographical areas. Following classifications in Geographic Information System (GIS) literature, FRIEDA targets all three categories of spatial relations: topological (border, equal, intersect, within), metric (distance), and directional (orientation). All questions require multi-step inference, and many require cross-map grounding and reasoning. We evaluate eleven state-of-the-art LVLMs under two settings: (1) the direct setting, where we provide the maps relevant to the question, and (2) the contextual setting, where the model may have to identify the maps relevant to the question before reasoning. Even the strongest models, Gemini-2.5-Pro and GPT-5-Think, achieve only 38.20% and 37.20% accuracy, respectively, far below human performance of 84.87%. These results reveal a persistent gap in multi-step cartographic reasoning, positioning FRIEDA as a rigorous benchmark to drive progress on spatial intelligence in LVLMs.

Correspondence as Video: Test-Time Adaption on SAM2 for Reference Segmentation in the Wild

Large vision models like the Segment Anything Model (SAM) exhibit significant limitations when applied to downstream tasks in the wild. Consequently, reference segmentation, which leverages reference images and their corresponding masks to impart novel knowledge to the model, emerges as a promising new direction for adapting vision models. However, existing reference segmentation approaches predominantly rely on meta-learning, which still necessitates an extensive meta-training process and brings massive data and computational cost. In this study, we propose a novel approach by representing the inherent correspondence between reference-target image pairs as a pseudo video. This perspective allows the latest version of SAM, known as SAM2, which is equipped with interactive video object segmentation (iVOS) capabilities, to be adapted to downstream tasks in a lightweight manner. We term this approach Correspondence As Video for SAM (CAV-SAM). CAV-SAM comprises two key modules: the Diffusion-Based Semantic Transition (DBST) module employs a diffusion model to construct a semantic transformation sequence, while the Test-Time Geometric Alignment (TTGA) module aligns the geometric changes within this sequence through test-time fine-tuning. We evaluated CAVSAM on widely-used datasets, achieving segmentation performance improvements exceeding 5% over SOTA methods. Implementation is provided in the supplementary materials.

GenHSI: Controllable Generation of Human-Scene Interaction Videos

Large-scale pre-trained video diffusion models have exhibited remarkable capabilities in diverse video generation. However, existing solutions face several challenges in using these models to generate long movie-like videos with rich human-object interactions that include unrealistic human-scene interaction, lack of subject identity preservation, and require expensive training. We propose GenHSI, a training-free method for controllable generation of long human-scene interaction videos (HSI). Taking inspiration from movie animation, our key insight is to overcome the limitations of previous work by subdividing the long video generation task into three stages: (1) script writing, (2) pre-visualization, and (3) animation. Given an image of a scene, a user description, and multiple images of a person, we use these three stages to generate long-videos that preserve human-identity and provide rich human-scene interactions. Script writing converts complex human tasks into simple atomic tasks that are used in the pre-visualization stage to generate 3D keyframes (storyboards). These 3D keyframes are rendered and animated by off-the-shelf video diffusion models for consistent long video generation with rich contacts in a 3D-aware manner. A key advantage of our work is that we alleviate the need for scanned, accurate scenes and create 3D keyframes from single-view images. We are the first to generate a long video sequence with a consistent camera pose that contains arbitrary numbers of character actions without training. Experiments demonstrate that our method can generate long videos that effectively preserve scene content and character identity with plausible human-scene interaction from a single image scene. Visit our project homepage https://kunkun0w0.github.io/project/GenHSI/ for more information.

T^2Agent A Tool-augmented Multimodal Misinformation Detection Agent with Monte Carlo Tree Search

Real-world multimodal misinformation often arises from mixed forgery sources, requiring dynamic reasoning and adaptive verification. However, existing methods mainly rely on static pipelines and limited tool usage, limiting their ability to handle such complexity and diversity. To address this challenge, we propose T2Agent, a novel misinformation detection agent that incorporates an extensible toolkit with Monte Carlo Tree Search (MCTS). The toolkit consists of modular tools such as web search, forgery detection, and consistency analysis. Each tool is described using standardized templates, enabling seamless integration and future expansion. To avoid inefficiency from using all tools simultaneously, a Bayesian optimization-based selector is proposed to identify a task-relevant subset. This subset then serves as the action space for MCTS to dynamically collect evidence and perform multi-source verification. To better align MCTS with the multi-source nature of misinformation detection, T2Agent extends traditional MCTS with multi-source verification, which decomposes the task into coordinated subtasks targeting different forgery sources. A dual reward mechanism containing a reasoning trajectory score and a confidence score is further proposed to encourage a balance between exploration across mixed forgery sources and exploitation for more reliable evidence. We conduct ablation studies to confirm the effectiveness of the tree search mechanism and tool usage. Extensive experiments further show that T2Agent consistently outperforms existing baselines on challenging mixed-source multimodal misinformation benchmarks, demonstrating its strong potential as a training-free approach for enhancing detection accuracy. The code will be released.

Video-SafetyBench: A Benchmark for Safety Evaluation of Video LVLMs

The increasing deployment of Large Vision-Language Models (LVLMs) raises safety concerns under potential malicious inputs. However, existing multimodal safety evaluations primarily focus on model vulnerabilities exposed by static image inputs, ignoring the temporal dynamics of video that may induce distinct safety risks. To bridge this gap, we introduce Video-SafetyBench, the first comprehensive benchmark designed to evaluate the safety of LVLMs under video-text attacks. It comprises 2,264 video-text pairs spanning 48 fine-grained unsafe categories, each pairing a synthesized video with either a harmful query, which contains explicit malice, or a benign query, which appears harmless but triggers harmful behavior when interpreted alongside the video. To generate semantically accurate videos for safety evaluation, we design a controllable pipeline that decomposes video semantics into subject images (what is shown) and motion text (how it moves), which jointly guide the synthesis of query-relevant videos. To effectively evaluate uncertain or borderline harmful outputs, we propose RJScore, a novel LLM-based metric that incorporates the confidence of judge models and human-aligned decision threshold calibration. Extensive experiments show that benign-query video composition achieves average attack success rates of 67.2%, revealing consistent vulnerabilities to video-induced attacks. We believe Video-SafetyBench will catalyze future research into video-based safety evaluation and defense strategies.

Rethinking the Role of Prompting Strategies in LLM Test-Time Scaling: A Perspective of Probability Theory

Recently, scaling test-time compute on Large Language Models (LLM) has garnered wide attention. However, there has been limited investigation of how various reasoning prompting strategies perform as scaling. In this paper, we focus on a standard and realistic scaling setting: majority voting. We systematically conduct experiments on 6 LLMs $\times$ 8 prompting strategies $\times$ 6 benchmarks. Experiment results consistently show that as the sampling time and computational overhead increase, complicated prompting strategies with superior initial performance gradually fall behind simple Chain-of-Thought. We analyze this phenomenon and provide theoretical proofs. Additionally, we propose a method according to probability theory to quickly and accurately predict the scaling performance and select the best strategy under large sampling times without extra resource-intensive inference in practice. It can serve as the test-time scaling law for majority voting. Furthermore, we introduce two ways derived from our theoretical analysis to significantly improve the scaling performance. We hope that our research can promote to re-examine the role of complicated prompting, unleash the potential of simple prompting strategies, and provide new insights for enhancing test-time scaling performance.

MacRAG: Compress, Slice, and Scale-up for Multi-Scale Adaptive Context RAG

Long-context (LC) Large Language Models (LLMs) combined with Retrieval-Augmented Generation (RAG) hold strong potential for complex multi-hop and large-document tasks. However, existing RAG systems often suffer from imprecise retrieval, incomplete context coverage under constrained context windows, and fragmented information caused by suboptimal context construction. We introduce Multi-scale Adaptive Context RAG (MacRAG), a hierarchical retrieval framework that compresses and partitions documents into coarse-to-fine granularities, then adaptively merges relevant contexts through chunk- and document-level expansions in real time. By starting from the finest-level retrieval and progressively incorporating higher-level and broader context, MacRAG constructs effective query-specific long contexts, optimizing both precision and coverage. Evaluations on the challenging LongBench expansions of HotpotQA, 2WikiMultihopQA, and Musique confirm that MacRAG consistently surpasses baseline RAG pipelines on single- and multi-step generation with Llama-3.1-8B, Gemini-1.5-pro, and GPT-4o. Our results establish MacRAG as an efficient, scalable solution for real-world long-context, multi-hop reasoning. Our code is available at https://github.com/Leezekun/MacRAG.

Art3D: Training-Free 3D Generation from Flat-Colored Illustration

Large-scale pre-trained image-to-3D generative models have exhibited remarkable capabilities in diverse shape generations. However, most of them struggle to synthesize plausible 3D assets when the reference image is flat-colored like hand drawings due to the lack of 3D illusion, which are often the most user-friendly input modalities in art content creation. To this end, we propose Art3D, a training-free method that can lift flat-colored 2D designs into 3D. By leveraging structural and semantic features with pre- trained 2D image generation models and a VLM-based realism evaluation, Art3D successfully enhances the three-dimensional illusion in reference images, thus simplifying the process of generating 3D from 2D, and proves adaptable to a wide range of painting styles. To benchmark the generalization performance of existing image-to-3D models on flat-colored images without 3D feeling, we collect a new dataset, Flat-2D, with over 100 samples. Experimental results demonstrate the performance and robustness of Art3D, exhibiting superior generalizable capacity and promising practical applicability. Our source code and dataset will be publicly available on our project page: https://joy-jy11.github.io/ .