Searcharxiv⌕ Search

arXiv subjects

Joey Tianyi Zhou

Publications and source records attributed to Joey Tianyi Zhou.

At least 37 records · Page 2Linked to original sources
arXiv2026

LaV-CoT: Language-Aware Visual CoT with Multi-Aspect Reward Optimization for Real-World Multilingual VQA

As large vision language models (VLMs) advance, their capabilities in multilingual visual question answering (mVQA) have significantly improved. Chain-of-thought (CoT) reasoning has been proven to enhance interpretability and complex reasoning. However, most existing approaches rely primarily on textual CoT and provide limited support for multilingual multimodal reasoning, constraining their deployment in real-world applications. To address this gap, we introduce LaV-CoT, the first Language-aware Visual CoT framework with Multi-Aspect Reward Optimization. LaV-CoT incorporates an interpretable multi-stage reasoning pipeline consisting of Text Summary with Bounding Box (BBox), Language Identification, Spatial Object-level Captioning, and Step-by-step Logical Reasoning. Following this reasoning pipeline, we design an automated data curation method that generates multilingual CoT annotations through iterative generation, correction, and refinement, enabling scalable and high-quality training data. To improve reasoning and generalization, LaV-CoT adopts a two-stage training paradigm combining Supervised Fine-Tuning (SFT) with Language-aware Group Relative Policy Optimization (GRPO), guided by verifiable multi-aspect rewards including language consistency, structural accuracy, and semantic alignment. Extensive evaluations on public datasets including MMMB, Multilingual MMBench, and MTVQA show that LaV-CoT achieves up to ~9.5% accuracy improvements over open-source baselines of similar size and even surpasses models with 2$\times$ larger scales by ~2.6%. Moreover, LaV-CoT outperforms advanced proprietary models such as GPT-4o-0513 and Gemini-2.5-flash. We further conducted an online A/B test to validate our method on real-world data, highlighting its effectiveness for industrial deployment. Our code is available at this link: https://github.com/HJNVR/LaV-CoT

cs.CV↗
arXiv2026

Reinforcement-Guided Synthetic Data Generation for Privacy-Sensitive Identity Recognition

High-fidelity generative models are increasingly needed in privacy-sensitive scenarios, where access to data is severely restricted due to regulatory and copyright constraints. This scarcity hampers model development--ironically, in settings where generative models are most needed to compensate for the lack of data. This creates a self-reinforcing challenge: limited data leads to poor generative models, which in turn fail to mitigate data scarcity. To break this cycle, we propose a reinforcement-guided synthetic data generation framework that adapts general-domain generative priors to privacy-sensitive identity recognition tasks. We first perform a cold-start adaptation to align a pretrained generator with the target domain, establishing semantic relevance and initial fidelity. Building on this foundation, we introduce a multi-objective reward that jointly optimizes semantic consistency, coverage diversity, and expression richness, guiding the generator to produce both realistic and task-effective samples. During downstream training, a dynamic sample selection mechanism further prioritizes high-utility synthetic samples, enabling adaptive data scaling and improved domain alignment. Extensive experiments on benchmark datasets demonstrate that our framework significantly improves both generation fidelity and classification accuracy, while also exhibiting strong generalization to novel categories in small-data regimes.

cs.CV↗
arXiv2026

Mitigating Entangled Steering in Large Vision-Language Models for Hallucination Reduction

Large Vision-Language Models (LVLMs) have achieved remarkable success across cross-modal tasks but remain hindered by hallucinations, producing textual outputs inconsistent with visual content. Existing methods mitigate hallucinations but often alter generation behavior, resulting in shorter outputs and shifted token distributions, especially in latent space steering approaches. We identify that this issue stems from entangled steering signals, where suppressing hallucinations inadvertently disrupts the model's intrinsic generation behavior. To address this, we propose MESA, an effective plug-and-play framework that performs controlled and selective latent intervention for hallucination mitigation. Specifically, MESA targets hallucination-relevant responses while preserving the model's original token distribution, enabling effective hallucination reduction without compromising generation behavior. Extensive experiments across diverse generative and discriminative benchmarks demonstrate that MESA consistently reduces hallucinations while better preserving generation behavior, outperforming prior methods across multiple LVLM families.

cs.CV↗
arXiv2026

AgentSwing: Adaptive Parallel Context Management Routing for Long-Horizon Web Agents

As large language models (LLMs) evolve into autonomous agents for long-horizon information-seeking, managing finite context capacity has become a critical bottleneck. Existing context management methods typically commit to a single fixed strategy throughout the entire trajectory. Such static designs may work well in some states, but they cannot adapt as the usefulness and reliability of the accumulated context evolve during long-horizon search. To formalize this challenge, we introduce a probabilistic framework that characterizes long-horizon success through two complementary dimensions: search efficiency and terminal precision. Building on this perspective, we propose AgentSwing, a state-aware adaptive parallel context management routing framework. At each trigger point, AgentSwing expands multiple context-managed branches in parallel and uses lookahead routing to select the most promising continuation. Experiments across diverse benchmarks and agent backbones show that AgentSwing consistently outperforms strong static context management methods, often matching or exceeding their performance with up to $3\times$ fewer interaction turns while also improving the ultimate performance ceiling of long-horizon web agents. Beyond the empirical gains, the proposed probabilistic framework provides a principled lens for analyzing and designing future context management strategies for long-horizon agents.

cs.CL↗
arXiv2026

SAG-Agent: Enabling Long-Horizon Reasoning in Strategy Games via Dynamic Knowledge Graphs

Most commodity software lacks accessible Application Programming Interfaces (APIs), requiring autonomous agents to interact solely through pixel-based Graphical User Interfaces (GUIs). In this API-free setting, large language model (LLM)-based agents face severe efficiency bottlenecks: limited to local visual experiences, they make myopic decisions and rely on inefficient trial-and-error, hindering both skill acquisition and long-horizon planning. To overcome these limitations, we propose SAG-Agent, an experience-driven learning framework that structures an agent's raw pixel-level interactions into a persistent State-Action Graph (SAG). SAG-Agent mitigates inefficient exploration by topologically linking functionally similar but visually distinct GUI states, constructing a rich neighborhood of experience that enables the agent to generalize from a diverse set of historical strategies. To facilitate long-horizon reasoning, we design a novel hybrid intrinsic reward mechanism based on the graph topology, combining a state-value reward for exploiting known high-value pathways with a novelty reward that encourages targeted exploration. This approach decouples strategic planning from pure discovery, allowing the agent to effectively value setup actions with delayed gratification. We evaluate SAG-Agent in two complex, open-ended GUI-based decision-making environments (Civilization V and Slay the Spire), demonstrating significant improvements in exploration efficiency and strategic depth over the state-of-the-art methods.

cs.AI↗
arXiv2026

IMS3: Breaking Distributional Aggregation in Diffusion-Based Dataset Distillation

Dataset Distillation aims to synthesize compact datasets that can approximate the training efficacy of large-scale real datasets, offering an efficient solution to the increasing computational demands of modern deep learning. Recently, diffusion-based dataset distillation methods have shown great promise by leveraging the strong generative capacity of diffusion models to produce diverse and structurally consistent samples. However, a fundamental goal misalignment persists: diffusion models are optimized for generative likelihood rather than discriminative utility, resulting in over-concentration in high-density regions and inadequate coverage of boundary samples crucial for classification. To address this issue, we propose two complementary strategies. Inversion-Matching (IM) introduces an inversion-guided fine-tuning process that aligns denoising trajectories with their inversion counterparts, broadening distributional coverage and enhancing diversity. Selective Subgroup Sampling(S^3) is a training-free sampling mechanism that improves inter-class separability by selecting synthetic subsets that are both representative and distinctive. Extensive experiments demonstrate that our approach significantly enhances the discriminative quality and generalization of distilled datasets, achieving state-of-the-art performance among diffusion-based methods.

cs.CV↗
arXiv2026

Agentic Spatio-Temporal Grounding via Collaborative Reasoning

Spatio-Temporal Video Grounding (STVG) aims to retrieve the spatio-temporal tube of a target object or person in a video given a text query. Most existing approaches perform frame-wise spatial localization within a predicted temporal span, resulting in redundant computation, heavy supervision requirements, and limited generalization. Weakly-supervised variants mitigate annotation costs but remain constrained by the dataset-level train-and-fit paradigm with an inferior performance. To address these challenges, we propose the Agentic Spatio-Temporal Grounder (ASTG) framework for the task of STVG towards an open-world and training-free scenario. Specifically, two specialized agents SRA (Spatial Reasoning Agent) and TRA (Temporal Reasoning Agent) constructed leveraging on modern Multimoal Large Language Models (MLLMs) work collaboratively to retrieve the target tube in an autonomous and self-guided manner. Following a propose-and-evaluation paradigm, ASTG duly decouples spatio-temporal reasoning and automates the tube extraction, verification and temporal localization processes. With a dedicate visual memory and dialogue context, the retrieval efficiency is significantly enhanced. Experiments on popular benchmarks demonstrate the superiority of the proposed approach where it outperforms existing weakly-supervised and zero-shot approaches by a margin and is comparable to some of the fully-supervised methods.

cs.CV↗
arXiv2026

SpecFLASH: A Latent-Guided Semi-autoregressive Speculative Decoding Framework for Efficient Multimodal Generation

Large language models and large multimodal models (LLMs and LMMs) deliver strong generative performance but suffer from slow decoding, a problem that becomes more severe when handling visual inputs, whose sequences typically contain many more tokens with lower information density than text. Speculative decoding accelerates LLM inference by letting a compact draft model propose candidate tokens that are selectively accepted by a larger target model, achieving speed-up without degrading quality. However, existing multimodal speculative decoding approaches largely ignore the structural characteristics of visual representations and usually rely on text-only draft models. In this paper, we introduce SpecFLASH, a speculative decoding framework tailored to LMMs that explicitly exploits multimodal structure when designing the draft model. We first mitigate redundancy in visual token sequences with a lightweight, latent-guided token compression module that compacts visual features while preserving semantics, and then leverage the co-occurrence and local correlations of visual entities via a semi-autoregressive decoding scheme that predicts multiple tokens in a single forward pass. Extensive experiments demonstrate that SpecFLASH consistently surpasses prior speculative decoding baselines, achieving up to $2.68\times$ speed-up on video captioning and $2.55\times$ on visual instruction tuning, relative to the original LMM. Our code is available here: https://github.com/ZihuaEvan/FlashSD/.

cs.CV↗
arXiv2026

SQLAgent: Learning to Explore Before Generating as a Data Engineer

Large Language Models have recently shown impressive capabilities in reasoning and code generation, making them promising tools for natural language interfaces to relational databases. However, existing approaches often fail to generalize in complex, real-world settings due to the highly database-specific nature of SQL reasoning, which requires deep familiarity with unique schemas, ambiguous semantics, and intricate join paths. To address this challenge, we introduce a novel two-stage LLM-based framework that decouples knowledge acquisition from query generation. In the Exploration Stage, the system autonomously constructs a database-specific knowledge base by navigating the schema with a Monte Carlo Tree Search-inspired strategy, generating triplets of schema fragments, executable queries, and natural language descriptions as usage examples. In the Deployment Stage, a dual-agent system leverages the collected knowledge as in-context examples to iteratively retrieve relevant information and generate accurate SQL queries in response to user questions. This design enables the agent to proactively familiarize itself with unseen databases and handle complex, multi-step reasoning. Extensive experiments on large-scale benchmarks demonstrate that our approach significantly improves accuracy over strong baselines, highlighting its effectiveness and generalizability.

cs.DB↗
arXiv2026

PandaPose: 3D Human Pose Lifting from a Single Image via Propagating 2D Pose Prior to 3D Anchor Space

3D human pose lifting from a single RGB image is a challenging task in 3D vision. Existing methods typically establish a direct joint-to-joint mapping from 2D to 3D poses based on 2D features. This formulation suffers from two fundamental limitations: inevitable error propagation from input predicted 2D pose to 3D predictions and inherent difficulties in handling self-occlusion cases. In this paper, we propose PandaPose, a 3D human pose lifting approach via propagating 2D pose prior to 3D anchor space as the unified intermediate representation. Specifically, our 3D anchor space comprises: (1) Joint-wise 3D anchors in the canonical coordinate system, providing accurate and robust priors to mitigate 2D pose estimation inaccuracies. (2) Depth-aware joint-wise feature lifting that hierarchically integrates depth information to resolve self-occlusion ambiguities. (3) The anchor-feature interaction decoder that incorporates 3D anchors with lifted features to generate unified anchor queries encapsulating joint-wise 3D anchor set, visual cues and geometric depth information. The anchor queries are further employed to facilitate anchor-to-joint ensemble prediction. Experiments on three well-established benchmarks (i.e., Human3.6M, MPI-INF-3DHP and 3DPW) demonstrate the superiority of our proposition. The substantial reduction in error by $14.7\%$ compared to SOTA methods on the challenging conditions of Human3.6M and qualitative comparisons further showcase the effectiveness and robustness of our approach.

cs.CV↗
arXiv2025

DP-CSGP: Differentially Private Stochastic Gradient Push with Compressed Communication

In this paper, we propose a Differentially Private Stochastic Gradient Push with Compressed communication (termed DP-CSGP) for decentralized learning over directed graphs. Different from existing works, the proposed algorithm is designed to maintain high model utility while ensuring both rigorous differential privacy (DP) guarantees and efficient communication. For general non-convex and smooth objective functions, we show that the proposed algorithm achieves a tight utility bound of $\mathcal{O}\left( \sqrt{d\log \left( \frac{1}δ \right)}/(\sqrt{n}Jε) \right)$ ($J$ and $d$ are the number of local samples and the dimension of decision variables, respectively) with $\left(ε, δ\right)$-DP guarantee for each node, matching that of decentralized counterparts with exact communication. Extensive experiments on benchmark tasks show that, under the same privacy budget, DP-CSGP achieves comparable model accuracy with significantly lower communication cost than existing decentralized counterparts with exact communication.

cs.LG↗
arXiv2025

Semantic Energy: Detecting LLM Hallucination Beyond Entropy

Large Language Models (LLMs) are being increasingly deployed in real-world applications, but they remain susceptible to hallucinations, which produce fluent yet incorrect responses and lead to erroneous decision-making. Uncertainty estimation is a feasible approach to detect such hallucinations. For example, semantic entropy estimates uncertainty by considering the semantic diversity across multiple sampled responses, thus identifying hallucinations. However, semantic entropy relies on post-softmax probabilities and fails to capture the model's inherent uncertainty, causing it to be ineffective in certain scenarios. To address this issue, we introduce Semantic Energy, a novel uncertainty estimation framework that leverages the inherent confidence of LLMs by operating directly on logits of penultimate layer. By combining semantic clustering with a Boltzmann-inspired energy distribution, our method better captures uncertainty in cases where semantic entropy fails. Experiments across multiple benchmarks show that Semantic Energy significantly improves hallucination detection and uncertainty estimation, offering more reliable signals for downstream applications such as hallucination detection.

cs.LG↗
arXiv2025

Cognitive Inception: Agentic Reasoning against Visual Deceptions by Injecting Skepticism

As the development of AI-generated contents (AIGC), multi-modal Large Language Models (LLM) struggle to identify generated visual inputs from real ones. Such shortcoming causes vulnerability against visual deceptions, where the models are deceived by generated contents, and the reliability of reasoning processes is jeopardized. Therefore, facing rapidly emerging generative models and diverse data distribution, it is of vital importance to improve LLMs' generalizable reasoning to verify the authenticity of visual inputs against potential deceptions. Inspired by human cognitive processes, we discovered that LLMs exhibit tendency of over-trusting the visual inputs, while injecting skepticism could significantly improve the models visual cognitive capability against visual deceptions. Based on this discovery, we propose \textbf{Inception}, a fully reasoning-based agentic reasoning framework to conduct generalizable authenticity verification by injecting skepticism, where LLMs' reasoning logic is iteratively enhanced between External Skeptic and Internal Skeptic agents. To the best of our knowledge, this is the first fully reasoning-based framework against AIGC visual deceptions. Our approach achieved a large margin of performance improvement over the strongest existing LLM baselines and SOTA performance on AEGIS benchmark.

cs.AI↗
arXiv2025

Low-Level Dataset Distillation for Medical Image Enhancement

Medical image enhancement is clinically valuable, but existing methods require large-scale datasets to learn complex pixel-level mappings. However, the substantial training and storage costs associated with these datasets hinder their practical deployment. While dataset distillation (DD) can alleviate these burdens, existing methods mainly target high-level tasks, where multiple samples share the same label. This many-to-one mapping allows distilled data to capture shared semantics and achieve information compression. In contrast, low-level tasks involve a many-to-many mapping that requires pixel-level fidelity, making low-level DD an underdetermined problem, as a small distilled dataset cannot fully constrain the dense pixel-level mappings. To address this, we propose the first low-level DD method for medical image enhancement. We first leverage anatomical similarities across patients to construct the shared anatomical prior based on a representative patient, which serves as the initialization for the distilled data of different patients. This prior is then personalized for each patient using a Structure-Preserving Personalized Generation (SPG) module, which integrates patient-specific anatomical information into the distilled dataset while preserving pixel-level fidelity. For different low-level tasks, the distilled data is used to construct task-specific high- and low-quality training pairs. Patient-specific knowledge is injected into the distilled data by aligning the gradients computed from networks trained on the distilled pairs with those from the corresponding patient's raw data. Notably, downstream users cannot access raw patient data. Instead, only a distilled dataset containing abstract training information is shared, which excludes patient-specific details and thus preserves privacy.

cs.CV↗
arXiv2025

Hulu-Med: A Transparent Generalist Model towards Holistic Medical Vision-Language Understanding

Real-world clinical decision-making requires integrating heterogeneous data, including medical text, 2D images, 3D volumes, and videos, while existing AI systems fail to unify all these signals, limiting their utility. In this paper, we introduce Hulu-Med, a transparent, generalist medical Vision-Language Model (VLM) designed to unify language-only, 2D/3D vision-language, and video understanding within a single architecture. Hulu-Med is trained on a curated corpus of 16.7 million samples, comprising exclusively public or synthetic data, spanning 12 major anatomical systems and 14 medical imaging modalities. Hulu-Med employs a medical-aware token-reduction strategy that prunes redundant visual tokens, achieving up to a 55% reduction for 3D and video inputs, improving cross-modal efficiency, and enabling training at 7B-32B parameter scales in approximately 4,000-40,000 GPU hours. Across 30 public in-domain and out-of-domain medical benchmarks-covering text reasoning, visual question answering, report generation, multilingual dialogue, video understanding, and rare disease diagnosis-Hulu-Med surpasses existing open-source models on 27 of 30 benchmarks and outperforms proprietary systems such as GPT-4o on 16 benchmarks. Despite being a VLM, Hulu-Med outperforms GPT-4o and matches GPT-o1 on the text-only HealthBench. For the first time in the community, we provide a fully transparent, reproducible and cost-effective pipeline for holistic medical vision-language understanding by releasing our end-to-end data curation, training procedures, and model parameters. Code and models are available at https://github.com/ZJUI-AI4H/Hulu-Med.

cs.CV↗
arXiv2025

SCOPE: Saliency-Coverage Oriented Token Pruning for Efficient Multimodel LLMs

Multimodal Large Language Models (MLLMs) typically process a large number of visual tokens, leading to considerable computational overhead, even though many of these tokens are redundant. Existing visual token pruning methods primarily focus on selecting the most salient tokens based on attention scores, resulting in the semantic incompleteness of the selected tokens. In this paper, we propose a novel visual token pruning strategy, called \textbf{S}aliency-\textbf{C}overage \textbf{O}riented token \textbf{P}runing for \textbf{E}fficient MLLMs (SCOPE), to jointly model both the saliency and coverage of the selected visual tokens to better preserve semantic completeness. Specifically, we introduce a set-coverage for a given set of selected tokens, computed based on the token relationships. We then define a token-coverage gain for each unselected token, quantifying how much additional coverage would be obtained by including it. By integrating the saliency score into the token-coverage gain, we propose our SCOPE score and iteratively select the token with the highest SCOPE score. We conduct extensive experiments on multiple vision-language understanding benchmarks using the LLaVA-1.5 and LLaVA-Next models. Experimental results demonstrate that our method consistently outperforms prior approaches. Our code is available at \href{https://github.com/kinredon/SCOPE}{https://github.com/kinredon/SCOPE}.

cs.CV↗
arXiv2025

Modest-Align: Data-Efficient Alignment for Vision-Language Models

Cross-modal alignment aims to map heterogeneous modalities into a shared latent space, as exemplified by models like CLIP, which benefit from large-scale image-text pretraining for strong recognition capabilities. However, when operating in resource-constrained settings with limited or low-quality data, these models often suffer from overconfidence and degraded performance due to the prevalence of ambiguous or weakly correlated image-text pairs. Current contrastive learning approaches, which rely on single positive pairs, further exacerbate this issue by reinforcing overconfidence on uncertain samples. To address these challenges, we propose Modest-Align, a lightweight alignment framework designed for robustness and efficiency. Our approach leverages two complementary strategies -- Random Perturbation, which introduces controlled noise to simulate uncertainty, and Embedding Smoothing, which calibrates similarity distributions in the embedding space. These mechanisms collectively reduce overconfidence and improve performance on noisy or weakly aligned samples. Extensive experiments across multiple benchmark datasets demonstrate that Modest-Align outperforms state-of-the-art methods in retrieval tasks, achieving competitive results with over 100x less training data and 600x less GPU time than CLIP. Our method offers a practical and scalable solution for cross-modal alignment in real-world, low-resource scenarios.

cs.CV↗
arXiv2025

DentVLM: A Multimodal Vision-Language Model for Comprehensive Dental Diagnosis and Enhanced Clinical Practice

Diagnosing and managing oral diseases necessitate advanced visual interpretation across diverse imaging modalities and integrated information synthesis. While current AI models excel at isolated tasks, they often fall short in addressing the complex, multimodal requirements of comprehensive clinical dental practice. Here we introduce DentVLM, a multimodal vision-language model engineered for expert-level oral disease diagnosis. DentVLM was developed using a comprehensive, large-scale, bilingual dataset of 110,447 images and 2.46 million visual question-answering (VQA) pairs. The model is capable of interpreting seven 2D oral imaging modalities across 36 diagnostic tasks, significantly outperforming leading proprietary and open-source models by 19.6% higher accuracy for oral diseases and 27.9% for malocclusions. In a clinical study involving 25 dentists, evaluating 1,946 patients and encompassing 3,105 QA pairs, DentVLM surpassed the diagnostic performance of 13 junior dentists on 21 of 36 tasks and exceeded that of 12 senior dentists on 12 of 36 tasks. When integrated into a collaborative workflow, DentVLM elevated junior dentists' performance to senior levels and reduced diagnostic time for all practitioners by 15-22%. Furthermore, DentVLM exhibited promising performance across three practical utility scenarios, including home-based dental health management, hospital-based intelligent diagnosis and multi-agent collaborative interaction. These findings establish DentVLM as a robust clinical decision support tool, poised to enhance primary dental care, mitigate provider-patient imbalances, and democratize access to specialized medical expertise within the field of dentistry.

cs.CV↗