arXiv subjects

Guangtao Zhai

Publications and source records attributed to Guangtao Zhai.

At least 19 recordsLinked to original sources

arXiv2026

KV Cache Quantization for Diffusion Language Models

KV cache quantization is well studied for autoregressive language models, where cached prefix states are exact under causal attention and quantization is mainly a cache reconstruction problem. Diffusion language models (dLLMs) break this assumption. Fast dLLM inference reuses KV states across denoising steps, but bidirectional token representations can change as masked tokens are revealed, so the reused cache may already differ from full recomputation before quantization is applied. We study KV cache quantization under this cache-reuse error. We adapt KIVI and KVQuant to Fast-dLLM-style prefix and dual cache reuse, and introduce TRACE and dLLM-KVQ. TRACE reconstructs feasible dLLM reveal trajectories from prompt-answer pairs, providing calibration states that match iterative masked decoding. dLLM-KVQ uses these trajectories to choose per-layer K-cache entries that remain in high precision according to output- and attention-level behavior metrics, while applying standard low-bit value quantization. Across LLaDA and Dream evaluations on GSM8K, HumanEval, MATH, and MBPP, dLLM-KVQ achieves the highest average score among quantized methods across the reported 4-bit and 2-bit prefix/dual cache settings.

cs.CL↗

arXiv2026

Efficient One-Step Diffusion Restoration Model with Compact Token Compression and Linear Attention

Real-world image super-resolution aims to recover high-quality images from complex and unknown real-world degradations. However, existing generative Real-ISR methods largely inherit the dense latent representations and quadratic-cost global modeling paradigm developed for high-resolution image synthesis, causing computation, memory usage, and inference latency to scale unfavorably with resolution and thus limiting practical deployment. We argue that the key bottleneck lies not in insufficient restoration priors, but in excessive token redundancy and costly token interactions during high-resolution restoration. Motivated by this observation, we revisit Real-ISR from the perspectives of compact latent representation and linear-complexity modeling, and propose SANA-SR, an efficient one-step restoration framework. Specifically, SANA-SR employs a deep compression autoencoder with a 32x compression ratio to drastically reduce latent tokens while preserving restoration-relevant structures and textures. On top of this compact latent space, we introduce a linear-attention DiT with LoRA fine-tuning, enabling efficient high-resolution restoration with linear-complexity token mixing. Extensive experiments on all benchmark datasets demonstrate that SANA-SR achieves highly competitive and often superior quantitative performance against existing methods, while restoring clearer and more realistic textures. Moreover, after pruning, the deployed model runs in 0.019s with 407.95G MACs and 344M parameters, highlighting its strong potential for practical mobile deployment.

cs.CV↗

arXiv2026

LG-Bench: A Graph-Structured Evaluation Benchmark for Life Sciences

Traditional evaluation benchmarks reduce inherently interconnected scientific knowledge in life sciences into flat lists of questions, disregarding the underlying topological structure of the knowledge. We introduce, the first graph-structured benchmark for life sciences, featuring over 10,000 high-quality multiple-choice questions across medicine, biology, and chemistry. Our approach constructs a weighted evaluation graph using bidirectional matching and semantic similarity algorithms, where nodes represent questions and edge weights capture their semantic relationships. Leveraging this graph topology, we design two novel evaluation metrics. The Global Coherence Score (GCS) measures a model's consistency within semantically related neighborhoods, while Knowledge Balance Score (KBS) analyzes how model errors are distributed across the graph to reveal conceptual blind spots. LG-Bench facilitates fine-grained comparison of LLMs by surfacing differences in conceptual coherence and patterns of knowledge organization across models. Our framework shifts the evaluation paradigm from flat accuracy metrics to structure-aware analysis, offering a new lens for diagnosing and improving LLM performance in the life sciences domain.

cs.AI↗

arXiv2026

DexBioLab: Benchmarking Dexterous Manipulation with Dynamic Experimental Phenomena in Realistic Laboratories

Automating biological laboratories requires robots to perform dexterous tool interactions under realistic visual conditions while handling dynamic experimental phenomena, presenting significant challenges for robotic manipulation. To enable systematic evaluation under these conditions, we introduce DexBioLab, a simulation benchmark and framework for dexterous manipulation in realistic biological laboratory environments. DexBioLab comprises 14 tasks across 4 robot embodiments, spanning functional tool use, precision manipulation, language-conditioned visual grounding, and instrument operation. We introduce controllable dynamic experimental phenomena, including flames, vapor, and bubbles, to capture the evolving visual conditions of laboratory procedures. To further enhance environmental realism, we develop a hybrid rendering pipeline that integrates interactive simulated foreground objects with Gaussian-splatting backgrounds reconstructed from 6 real laboratory scenes. For efficient large-scale data generation and evaluation, we further build a parallel simulation framework on top of mjlab, proposing Parallel Finite-State Machine (PFSM) and integrating batched rendering with decoupled trajectory replay for demonstration generation and model-independent closed-loop policy evaluation. Our evaluation reveals substantial variation in policy performance across tasks, exposing distinct strengths and weaknesses of current manipulation policies. Real-world experiments further demonstrate a reduced sim-to-real gap, supporting the validity of AutoBio+ for realistic manipulation evaluation. Supplementary materials are available at https://cumulusalpha.github.io/DexBioLabWebPage.

cs.RO↗

arXiv2026

Trapped in Your Story: Evaluating Narrative-Induced Sycophancy in Large Language Models

Large language models (LLMs) are known to exhibit sycophancy by adjusting their responses toward users' expressed views. However, real users often justify their opinions through reasons, personal experiences, causal explanations, and broader narratives, raising the question of whether such narratives can further steer model judgments. We construct a controlled benchmark of 500 debatable questions across ten domains, where the user stance is held fixed while its justification is systematically varied without introducing new task-relevant evidence. Each question is instantiated under 11 matched conditions: one neutral baseline and five mirrored narrative levels for each of two opposing positions, yielding 5,500 evaluation instances. We evaluate seven LLMs and observe substantial differences in narrative-induced agreement. The effect is not monotonic with narrative elaboration: personal experience can reduce agreement, whereas causal framing and worldview-based synthesis increase it. Controlled GPT-5.1 ablations further show that causal framing produces the strongest positive effect, while claim-preserving bullet presentation also increases agreement over paragraph form. These findings suggest that sycophancy is shaped not only by user stance, but also by how supporting explanations are structured and presented.

cs.AI↗

arXiv2026

Robo2Process: Benchmarking and Learning Process-Aware Understanding in Vision-Language Robotic Manipulation

Vision-language models (VLMs) used as robotic critics, reward generators, and failure detectors must understand how manipulation unfolds, not only whether it succeeds. Yet fine-grained process supervision is not readily available in a unified form, and existing evaluations do not systematically diagnose these abilities. We introduce Robo2Process, a trace-grounded framework for scalable construction of process-aware supervision from heterogeneous robotic execution data. It aligns visual observations with source-native signals and annotations, extracts local process units (LPUs), and generates structured QA supervision through task-specific sampling, deterministic label construction, and ambiguity filtering. We instantiate this framework as RoboProcessBench, a diagnostic benchmark with 12 question families spanning \emph{static monitoring} and \emph{dynamic reasoning}, and ProcessData, a corpus of 57,892 QA pairs across four data sources and 260 manipulation tasks, split into ProcessData-SFT and ProcessData-Eval. Evaluation of a broad suite of VLMs reveals uneven process understanding. Supervision from ProcessData-SFT improves \textit{Qwen2.5-VL-7B} and \textit{InternVL-3-8B} on local state, motion, progress, and primitive-aware cues. External evaluation on RoboRewardBench shows Robo2Process enables out-of-distribution process-aware reasoning abilities. Overall, Robo2Process provides a source-aware pipeline, diagnostic benchmark, and learnable substrate for process-aware understanding in robotic manipulation.

cs.RO↗

arXiv2026

AT-ADD: A Benchmark and Challenge for Robust and All-Type Audio Deepfake Detection

Recent audio generation models can synthesize high-fidelity speech, environmental sound, singing voice, and music, creating new risks for multimedia trust. Existing audio deepfake detection (ADD) benchmarks remain predominantly speech-centric and often under-represent realistic channel variation and heterogeneous audio types. This paper presents all-type ADD (AT-ADD), a large-scale benchmark and challenge designed to evaluate both robust speech deepfake detection and all-type audio deepfake detection. Track 1 evaluates speech deepfake detection under unseen synthesizers, diverse recording conditions, signal perturbations, and replay effects. Track 2 evaluates type-agnostic deepfake detection over speech, sound, singing, and music when the audio type is unknown at test time. We detail the dataset construction, evaluation protocol, and reproducible baselines, and analyze the final systems submitted to the ACM Multimedia 2026 Grand Challenge. The strongest official baseline obtains 76.73% and 79.47% Macro-F1 on the Track 1 and Track 2 evaluation sets, respectively, whereas the winning challenge systems reach 90.71% and 96.10%. Beyond the aggregate rankings, sample-level analysis of the top five submissions reveals concentrated failures on specific synthesis methods and replayed speech. It also quantifies cross-system error complementarity and separates stable leaderboard gaps from statistically ambiguous adjacent ranks. Our results show that large-scale self-supervised representations, condition-aware augmentation, multi-crop inference, and structured fusion or routing are central to generalization, while generator-specific robustness and balanced detection across heterogeneous audio types remain unresolved.

cs.SD↗

arXiv2026

ManipShield: A Unified Framework for Image Manipulation Detection, Localization and Explanation

With the rapid advancement of generative models, powerful image editing methods now enable diverse and highly realistic image manipulations that far surpass traditional deepfake techniques, posing new challenges for manipulation detection. Existing image manipulation detection and localization (IMDL) benchmarks suffer from limited content diversity, narrow generative-model coverage, and insufficient interpretability, which hinders the generalization and explanation capabilities of current manipulation detection methods. To address these limitations, we introduce \textbf{ManipBench}, a large-scale benchmark for image manipulation detection and localization focusing on AI-edited images. ManipBench contains over 450K manipulated images produced by 25 state-of-the-art image editing models across 12 manipulation categories, among which 100K images are further annotated with bounding boxes, judgment cues, and textual explanations to support interpretable detection. Building upon ManipBench, we propose \textbf{ManipShield}, an all-in-one model based on a Multimodal Large Language Model (MLLM) that leverages contrastive LoRA fine-tuning and task-specific decoders to achieve unified image manipulation detection, localization, and explanation. Extensive experiments on ManipBench and several public datasets demonstrate that ManipShield achieves state-of-the-art performance and exhibits strong generality to unseen manipulation models. The project is available at https://github.com/IntMeGroup/ManipShield.

cs.CV↗

arXiv2026

EmoGenie: Benchmarking and Evaluating Expressed and Evoked Emotion in T2I Generation with MLLMs

Rapid advances in artificial intelligence-generated content (AIGC) have significantly promoted using text-to-image (T2I) models to generate images with specific affective intentions. However, the emotional effects conveyed by generated images frequently deviate from target expectations, and it remains unclear whether multimodal large language models (MLLMs) can reliably interpret the emotion of AI-generated visual content. Existing affective benchmarks are predominantly built on natural images, while affective studies of generated images have largely focused on emotions expressed in portraits, lacking a unified benchmark for studying the emotions expressed and evoked by generated images. We therefore introduce EmoGenie, a benchmark of emotion for AI-generated images that evaluates capabilities in T2I generation and MLLM understanding across expressed and evoked emotions. It comprises 6,400 images generated by eight T2I models using 800 emotion-targeted prompts, annotated with question-answer (QA) pairs, termed EmoGenie-QA, and continuous human ratings, termed EmoGenie-Assess. EmoGenie-QA contains 90,000 closed-set questions in single-image and image-pair settings, as well as 1,057 open-ended questions for emotion description. EmoGenie-Assess benchmarks continuous affective and aesthetic prediction using 13,175 mean opinion scores (MOSs) aggregated from human ratings in terms of valence, arousal, dominance, and aesthetic quality. We further develop EmoProbe, which combines emotion question answering with response-aligned, dimension-specific regression heads for continuous score prediction. Experimental results show that EmoProbe achieves superior results on emotion-related QA tasks, while its affective and aesthetic predictions correlate better with human ratings, demonstrating its state-of-the-art capabilities of unified coarse-to-fine affective understanding for AI-generated images.

cs.CV↗

arXiv2026

Q-Save: Towards Scoring and Attribution for Generated Video Evaluation

Evaluating AI-generated video (AIGV) quality hinges on three crucial dimensions: visual quality, dynamic quality, and text-video alignment. While numerous evaluation datasets and algorithms have been proposed, existing approaches are constrained by two limitations: the absence of systematic definitions for evaluation dimensions, and the isolated treatment of the three dimensions in separate models. Therefore, we introduce Q-Save, a holistic benchmark dataset and unified evaluation model for AIGV quality assessment. The Q-Save dataset contains nearly 10,000 video samples, each annotated with Mean Opinion Scores (MOS) and fine-grained attribution explanations across the three core dimensions. Leveraging this attribution-annotated dataset, we train the proposed Q-Save model, which adopts the SlowFast framework to balance accuracy and efficiency, and employs a three-stage training strategy with Chain-of-Thought (COT) formatted data: Supervised Fine-Tuning (SFT), Grouped Relative Policy Optimization (GRPO), and a final SFT round for stability, to jointly perform quality scoring and attribution generation. Experimental results demonstrate that Q-Save achieves superior performance in AIGV quality prediction while providing interpretable justifications. Code and dataset will be released upon publication.

cs.CV↗

arXiv2026

Constrained Dynamic Gaussian Splatting

While Dynamic Gaussian Splatting enables high-fidelity 4D reconstruction, its deployment is severely hindered by a fundamental dilemma: unconstrained densification leads to excessive memory consumption incompatible with edge devices, whereas heuristic pruning fails to achieve optimal rendering quality under preset Gaussian budgets. In this work, we propose Constrained Dynamic Gaussian Splatting (CDGS), a novel framework that formulates dynamic scene reconstruction as a budget-constrained optimization problem to enforce a strict, user-defined Gaussian budget during training. Our key insight is to introduce a differentiable budget controller as the core optimization driver. Guided by a multi-modal unified importance score, this controller fuses geometric, motion, and perceptual cues for precise capacity regulation. To maximize the utility of this fixed budget, we further introduce an adaptive static-dynamic allocation strategy that separates the Gaussian representation into static and dynamic branches and distributes the shared global capacity between them according to motion complexity. Furthermore, we implement a three-phase training strategy to seamlessly integrate these constraints, ensuring precise adherence to the target count. After training, a dual-mode hybrid compression scheme further reduces storage overhead. CDGS therefore not only strictly adheres to the specified Gaussian-count budget (error<2%) but also achieves favorable rate-distortion performance. Extensive experiments demonstrate that CDGS delivers optimal rendering quality under varying capacity limits and favorable rate-distortion performance, achieving over 3x model compression compared with the state-of-the-art method Ex4DGS.

cs.CV↗

arXiv2026

From Pixel to Coding: Evaluating the Figure Reproduction Capabilities of MLLMs

Multimodal Large Language Models (MLLMs) have demonstrated impressive capabilities in both visual understanding and code generation. However, existing benchmarks typically evaluate these two modalities in isolation, lacking a dedicated assessment of their unification, i.e., how a model can perceive complex visual structures and synthesize them into precise, executable code. Moreover, current visual code generation benchmarks often rely on simplified layouts within single programming environments, falling short of evaluating true unified multimodal reasoning. To bridge this gap, we propose FigCodeBench, a comprehensive framework for rigorously evaluating MLLMs on figure reproduction, integrating multimodal comprehension and generation. We first design a systematic dataset construction pipeline, resulting in a total of 6,194 instances that cover 7 functional categories and 4 types of programming languages. We further categorize figure reproduction into three tiers with visual and code complexity modeling, specifically targeting complex structural reasoning, varying aspect ratios, and dense geometric constraints. We introduce a multi-dimensional evaluation protocol, encompassing visual fidelity and syntactic isomorphism, that aligns highly with the Mean Machine Opinion Score (MMOS) and human preferences. Based on our framework, we conducted extensive experiments on 24 widely used proprietary and open-source MLLMs (e.g., Gemini 3.1 Pro, GPT-5.4, and Kimi-K2.5), where we observed a universal, non-linear performance cliff across different programming languages and difficulty scenarios for all models, and gained several insights, such as the significant metric decline in rigid declarative languages.

cs.CV↗

arXiv2026

Harnessing Multimodal Large Language Models for Training-Free Human-Object Interaction Detection

Human-object interaction (HOI) detection aims to localize human-object pairs and recognize their interactions. Traditional supervised methods perform strongly but rely on task-specific training. Recent multimodal large language models (MLLMs) offer a promising route to training-free HOI detection through their broad visual-semantic knowledge and versatile perceptual and reasoning capabilities. However, existing approaches largely invoke these capabilities through loosely coordinated inference stages. This fragmented execution restricts the role of interaction hypotheses in guiding visual exploration, leaving key participants overlooked and local ambiguities unresolved. Furthermore, propagating early semantic assumptions through subsequent visual grounding and relation prediction induces self-reinforcing semantic circularity. To resolve these challenges, we propose HarnessHOI, a training-free framework that transforms passive MLLM inference into an active interaction-centric harness. Specifically, we introduce an interaction-guided perception mechanism that projects emerging interaction hypotheses back into the visual space to discover missing participants and refine ambiguous evidence through targeted observation. Furthermore, a relation-agnostic geometric adjudication module reconciles multi-source evidence to establish a unified spatial basis for grounded interaction reasoning across multiple actions and semantic roles. Extensive experiments on HICO-DET and V-COCO demonstrate that HarnessHOI achieves state-of-the-art performance among training-free methods, confirming the effectiveness of the proposed harness for complex interaction understanding. Code will be released upon publication.

cs.CV↗

arXiv2026

Evaluating the Evaluators: Diagnosing Large Multimodal Models for AI-Generated Image Assessment

With the rapid advancement of text-to-image (T2I) generation, robust evaluation becomes critical yet challenging, as traditional metrics fail to capture fine-grained alignment and generative artifacts. While large multimodal models (LMMs) are increasingly adopted as evaluators, existing benchmarks typically study semantic understanding, quality perception, and authenticity identification in isolation, while largely neglecting responsibility detection. This leaves a gap in unified and comprehensive validation. To bridge this gap, we introduce SQUARE-Bench, a comprehensive benchmark that systematically evaluates LMM capabilities as evaluators of AI-generated images across four aspects: Semantics, Quality, Authenticity, and Responsibility. SQUARE-Bench introduces a granular taxonomy of 38 sub-dimensions to evaluate nearly 10K AI-generated images sampled from 22 diverse models, ranging from legacy to state-of-the-art generators, complemented by over 3K real-world images. The images are annotated with curated question-answering pairs. Extensive experiments on 23 LMMs reveal that top proprietary models, such as Gemini-3-Pro, already outperform the individual human expert baseline. However, the performance gap between models remains significant, exhibiting notable disparities in fine-grained inference and domain-specific robustness. Beyond benchmarking, we conduct a proof-of-concept study of LMM-guided iterative editing, in which dimension-specific LMMs provide diagnostic feedback to fixed image editors. The resulting guided system yields selective improvements in semantics, authenticity, and responsibility, while exhibiting a consistent visual-quality trade-off. SQUARE-Bench can serve as both a diagnostic tool for characterizing LMM evaluator capabilities and studying their use in T2I generation refinement. The benchmark and dataset will be released upon publication.

cs.CV↗

arXiv2026

Invisible in Space, Visible in Time: Motion Vision CAPTCHA against GUI Agents

Most existing visual CAPTCHAs remain spatially solvable: the required information is exposed by static appearance, local structure, and interface state. This assumption is weakened by advances in multimodal large language models (MLLMs) and Graphical User Interface (GUI) agents, which exhibit strong visual perception, reasoning, and browser interaction capabilities. We propose Motion Vision CAPTCHA (MVCAP), a hierarchical motion-based CAPTCHA framework in which target semantics are instantiated as motion-defined foreground structures and become recoverable only through temporal segregation from a dynamically evolving background. Built on this shared principle, MVCAP is instantiated in three perceptually progressive levels: coherent motion, structural motion, and biological motion. To evaluate this framework, we introduce MVCAP-Bench, a browser-based benchmark with 600 live CAPTCHA instances, together with a matched foreground-only control benchmark, MVCAP-Bench-FG. We evaluate humans, Browser Use agents, native computer use agents, and a supplementary offline VQA setting derived from the same instances. Results reveal a substantial human--agent gap: on the full MVCAP-Bench, human accuracy reaches 99.6%, whereas the best GUI agent achieves only 16.8%, close to the six-way chance level. The foreground-only control further shows that the key difficulty comes from dynamic background camouflage rather than answer format or browser interaction alone. These findings identify a measurable human--agent perception gap and position MVCAP-Bench as a benchmark for studying motion-defined perception in current agents.

cs.CV↗

arXiv2026

SafetyFlow: An Agent-Flow System for Automated LLM Safety Benchmarking

The rapid proliferation of large language models (LLMs) has intensified the requirement for reliable safety evaluation to uncover model vulnerabilities. To this end, numerous LLM safety evaluation benchmarks are proposed. However, existing benchmarks generally rely on labor-intensive manual curation, which causes excessive time and resource consumption. They also exhibit significant redundancy and limited difficulty. To alleviate these problems, we introduce SafetyFlow, the first agent-flow system designed to automate the construction of LLM safety benchmarks. SafetyFlow can automatically build a comprehensive safety benchmark in only four days without any human intervention by orchestrating seven specialized agents, significantly reducing time and resource cost. Equipped with versatile tools, the agents of SafetyFlow ensure process and cost controllability while integrating human expertise into the automatic pipeline. The final constructed dataset, SafetyFlowBench, contains 23,446 queries with low redundancy and strong discriminative power. Our contribution includes the first fully automated benchmarking pipeline and a comprehensive safety benchmark. We evaluate the safety of 49 advanced LLMs on our dataset and conduct extensive experiments to validate our efficacy and efficiency.

cs.CL↗

arXiv2026

LM Fight Arena: Benchmarking Large Multimodal Models via Game Competition

Existing benchmarks for large multimodal models (LMMs) often fail to capture their performance in real-time, adversarial environments. We introduce LM Fight Arena (Large Model Fight Arena), a novel framework that evaluates LMMs by pitting them against each other in the classic fighting game Mortal Kombat II, a task requiring rapid visual understanding and tactical, sequential decision-making. In a controlled tournament, we test six leading open- and closed-source models, where each agent operates controlling the same character to ensure a fair comparison. The models are prompted to interpret game frames and state data to select their next actions. Unlike static evaluations, LM Fight Arena provides a fully automated, reproducible, and objective assessment of an LMM's strategic reasoning capabilities in a dynamic setting. This work introduces a challenging and engaging benchmark that bridges the gap between AI evaluation and interactive entertainment.

cs.AI↗

arXiv2026

Q-SiT: Teaching LMMs for Image Quality Scoring and Interpreting

Image quality scoring and interpreting are two fundamental components of Image Quality Assessment (IQA). The former quantifies image quality, while the latter enables descriptive question answering about image quality. Traditionally, these two tasks have been addressed independently. However, image-quality-specific psychophysical studies suggest that these two tasks are conceptually interconnected: interpreting explicitly represents perceived quality attributes whereas scoring summarizes such evidence into an overall quality judgment. Thus, unifying these capabilities within a single model is both intuitive and logically coherent. In this paper, we propose Q-SiT (Quality Scoring and Interpreting joint Teaching), a unified framework that enables large multimodal models (LMMs) to learn both image quality scoring and interpreting simultaneously. We achieve this by transforming conventional IQA datasets into learnable question-answering datasets and incorporating human-annotated quality interpreting data for training. Furthermore, we introduce an efficient scoring \& interpreting balance strategy, which first determines the optimal data mix ratio on lightweight LMMs and then maps this ratio to primary LMMs for fine-tuning adjustment. This strategy not only mitigates task interference and enhances cross-task knowledge transfer but also significantly reduces computational costs compared to direct optimization on full-scale LMMs. With this joint learning framework and corresponding training strategy, we develop Q-SiT, the first model capable of simultaneously performing image quality scoring and interpreting tasks, along with its lightweight variant, Q-SiT-mini. Experimental results demonstrate that Q-SiT achieves strong performance in both tasks with superior generalization IQA abilities, while Q-SiT-mini significantly reduces computational overhead while maintaining competitive performance.

cs.CV↗