Yuchen Yuan

Yuxiang He, Jian Zhao, Yuchen Yuan et al.

The exponential growth of digital content presents significant challenges for content safety. Current moderation systems, often based on single models or fixed pipelines, exhibit limitations in identifying implicit risks and providing interpretable judgment processes. To address these issues, we propose Aetheria, a multimodal interpretable content safety framework based on multi-agent debate and collaboration.Employing a collaborative architecture of five core agents, Aetheria conducts in-depth analysis and adjudication of multimodal content through a dynamic, mutually persuasive debate mechanism, which is grounded by RAG-based knowledge retrieval.Comprehensive experiments on our proposed benchmark (AIR-Bench) validate that Aetheria not only generates detailed and traceable audit reports but also demonstrates significant advantages over baselines in overall content safety accuracy, especially in the identification of implicit risks. This framework establishes a transparent and interpretable paradigm, significantly advancing the field of trustworthy AI content moderation.

Yuchen Yuan, Junhuan Yang, Hao Wan et al.

Scientific machine learning (SciML) is increasingly applied to in-field processing, controlling, and monitoring; however, wide-area sensing, real-time demands, and strict energy and reliability constraints make centralized SciML implementation impractical. Most SciML models assume raw data aggregation at a central node, incurring prohibitively high communication latency and energy costs; yet, distributing models developed for general-purpose ML often breaks essential physical principles, resulting in degraded performance. To address these challenges, we introduce EPIC, a hardware- and physics-co-guided distributed SciML framework, using full-waveform inversion (FWI) as a representative task. EPIC performs lightweight local encoding on end devices and physics-aware decoding at a central node. By transmitting compact latent features rather than high-volume raw data and by using cross-attention to capture inter-receiver wavefield coupling, EPIC significantly reduces communication cost while preserving physical fidelity. Evaluated on a distributed testbed with five end devices and one central node, and across 10 datasets from OpenFWI, EPIC reduces latency by 8.9$\times$ and communication energy by 33.8$\times$, while even improving reconstruction fidelity on 8 out of 10 datasets.

Wei Cai, Shujuan Liu, Jian Zhao et al.

Multimodal Large Language Models (MLLMs) are susceptible to the implicit reasoning risk, wherein innocuous unimodal inputs synergistically assemble into risky multimodal data that produce harmful outputs. We attribute this vulnerability to the difficulty of MLLMs maintaining safety alignment through long-chain reasoning. To address this issue, we introduce Safe-Semantics-but-Unsafe-Interpretation (SSUI), the first dataset featuring interpretable reasoning paths tailored for such a cross-modal challenge. A novel training framework, Safety-aware Reasoning Path Optimization (SRPO), is also designed based on the SSUI dataset to align the MLLM's internal reasoning process with human safety values. Experimental results show that our SRPO-trained models achieve state-of-the-art results on key safety benchmarks, including the proposed Reasoning Path Benchmark (RSBench), significantly outperforming both open-source and top-tier commercial MLLMs.

Huilin Zhou, Jian Zhao, Yilu Zhong et al.

Red teaming is critical for uncovering vulnerabilities in Large Language Models (LLMs). While automated methods have improved scalability, existing approaches often rely on static heuristics or stochastic search, rendering them brittle against advanced safety alignment. To address this, we introduce Metis, a framework that reformulates jailbreaking as inference-time policy optimization within an adversarial Partially Observable Markov Decision Process (POMDP). Metis employs a self-evolving metacognitive loop to perform causal diagnosis of a target's defense logic and leverages structured feedback as a semantic gradient to refine its policy, offering enhanced interpretability through transparent reasoning traces. Extensive evaluations across 10 diverse models demonstrate that Metis achieves the strongest average Attack Success Rate (ASR) among compared methods at 89.2%, maintaining high efficacy on resilient frontier models (e.g., 76.0% on O1 and 78.0% on GPT-5-chat) where traditional baselines exhibit substantial performance degradation. By replacing redundant exploration with directed optimization, Metis reduces token costs by an average of 8.2x and up to 11.4x. Our analysis reveals that current defenses remain vulnerable to internally-steered, closed-loop reasoning trajectories under the tested settings, highlighting a critical need for next-generation defenses capable of reasoning about safety dynamically during inference.

Wei Cai, Jian Zhao, Yuchen Yuan et al.

Multimodal Large Language Models (MLLMs), despite their advances, are hindered by their high hallucination tendency and heavy reliance on brittle, linear reasoning processes, leading to failures in complex tasks. To address these limitations, we introduce Visual Attention Reasoning (VAR), a novel framework that recasts grounded reasoning as a structured search over a reasoning trajectory space. VAR decomposes the reasoning process into two key stages: traceable evidence grounding and search-based chain-of-thought (CoT) generation, which incorporates a backtracking mechanism for self-correction. The search is guided by a multi-faceted reward function with semantic and geometric self-verification components, which penalize outputs that are not faithfully grounded in the visual input. We provide a theoretical analysis for our search strategy, validating its capability to find the correct solution with high probability. Experimental results show that our 7B model, VAR-7B, sets a new state-of-the-art on a comprehensive suite of hallucination and safety benchmarks, significantly outperforming existing open-source models and demonstrating competitive performance against leading proprietary systems.

Qiao Yan, Yuchen Yuan, Xiaowei Hu et al.

The increasing use of vision-language models (VLMs) in healthcare applications presents great challenges related to hallucinations, in which the models may generate seemingly plausible results that are in fact incorrect. Such hallucinations can jeopardize clinical decision making, potentially harming the diagnosis and treatments. In this work, we propose MedHallTune, a large-scale benchmark designed specifically to evaluate and mitigate hallucinations in medical VLMs. Comprising over 100,000 images and 1,000,000 instruction pairs, MedHallTune includes both hallucination and non-hallucination samples, each with ground-truth annotations. We conduct a comprehensive evaluation of current medical and general VLMs using MedHallTune, assessing their performance across key metrics, including clinical accuracy, relevance, detail level, and risk level. The experimental results show that fine-tuning with MedHallTune successfully improves the ability of several existing models to manage hallucinations and boost their zero-shot performance on downstream visual-question-answering (VQA) tasks, making them more reliable for practical medical applications. Our work contributes to the development of more trustworthy VLMs. Codes and dataset will be available at \href{https://github.com/russellyq/MedHallTune}{MedHallTune}.

Xiangyun Zhao, Yi Yang, Feng Zhou et al.

Recognizing attributes of objects and their parts is important to many computer vision applications. Although great progress has been made to apply object-level recognition, recognizing the attributes of parts remains less applicable since the training data for part attributes recognition is usually scarce especially for internet-scale applications. Furthermore, most existing part attribute recognition methods rely on the part annotation which is more expensive to obtain. To solve the data insufficiency problem and get rid of dependence on the part annotation, we introduce a novel Concept Sharing Network (CSN) for part attribute recognition. A great advantage of CSN is its capability of recognizing the part attribute (a combination of part location and appearance pattern) that has insufficient or zero training data, by learning the part location and appearance pattern respectively from the training data that usually mix them in a single label. Extensive experiments on CUB-200-2011 [51], CelebA [35] and a newly proposed human attribute dataset demonstrate the effectiveness of CSN and its advantages over other methods, especially for the attributes with few training samples. Further experiments show that CSN can also perform zero-shot part attribute recognition. The code will be made available at https://github.com/Zhaoxiangyun/Concept-Sharing-Network.

Kaiyu Yue, Ming Sun, Yuchen Yuan et al.

The non-local module is designed for capturing long-range spatio-temporal dependencies in images and videos. Although having shown excellent performance, it lacks the mechanism to model the interactions between positions across channels, which are of vital importance in recognizing fine-grained objects and actions. To address this limitation, we generalize the non-local module and take the correlations between the positions of any two channels into account. This extension utilizes the compact representation for multiple kernel functions with Taylor expansion that makes the generalized non-local module in a fast and low-complexity computation flow. Moreover, we implement our generalized non-local method within channel groups to ease the optimization. Experimental results illustrate the clear-cut improvements and practical applicability of the generalized non-local module on both fine-grained object recognition and video classification. Code is available at: https://github.com/KaiyuYue/cgnl-network.pytorch.

Xikai Yang, Juzheng Miao, Yuchen Yuan et al.

Medical large vision-language models (LVLMs) have demonstrated promising performance across various single-image question answering (QA) benchmarks, yet their capability in processing multi-image clinical scenarios remains underexplored. Unlike single image based tasks, medical tasks involving multiple images often demand sophisticated visual understanding capabilities, such as temporal reasoning and cross-modal analysis, which are poorly supported by current medical LVLMs. To bridge this critical gap, we present the Med-MIM instruction dataset, comprising 83.2K medical multi-image QA pairs that span four types of multi-image visual abilities (temporal understanding, reasoning, comparison, co-reference). Using this dataset, we fine-tune Mantis and LLaVA-Med, resulting in two specialized medical VLMs: MIM-LLaVA-Med and Med-Mantis, both optimized for multi-image analysis. Additionally, we develop the Med-MIM benchmark to comprehensively evaluate the medical multi-image understanding capabilities of LVLMs. We assess eight popular LVLMs, including our two models, on the Med-MIM benchmark. Experimental results show that both Med-Mantis and MIM-LLaVA-Med achieve superior performance on the held-in and held-out subsets of the Med-MIM benchmark, demonstrating that the Med-MIM instruction dataset effectively enhances LVLMs' multi-image understanding capabilities in the medical domain.

Xikai Yang, Jian Wu, Xi Wang et al.

Glaucoma is one of the major eye diseases that leads to progressive optic nerve fiber damage and irreversible blindness, afflicting millions of individuals. Glaucoma forecast is a good solution to early screening and intervention of potential patients, which is helpful to prevent further deterioration of the disease. It leverages a series of historical fundus images of an eye and forecasts the likelihood of glaucoma occurrence in the future. However, the irregular sampling nature and the imbalanced class distribution are two challenges in the development of disease forecasting approaches. To this end, we introduce the Multi-scale Spatio-temporal Transformer Network (MST-former) based on the transformer architecture tailored for sequential image inputs, which can effectively learn representative semantic information from sequential images on both temporal and spatial dimensions. Specifically, we employ a multi-scale structure to extract features at various resolutions, which can largely exploit rich spatial information encoded in each image. Besides, we design a time distance matrix to scale time attention in a non-linear manner, which could effectively deal with the irregularly sampled data. Furthermore, we introduce a temperature-controlled Balanced Softmax Cross-entropy loss to address the class imbalance issue. Extensive experiments on the Sequential fundus Images for Glaucoma Forecast (SIGF) dataset demonstrate the superiority of the proposed MST-former method, achieving an AUC of 98.6% for glaucoma forecasting. Besides, our method shows excellent generalization capability on the Alzheimer's Disease Neuroimaging Initiative (ADNI) MRI dataset, with an accuracy of 90.3% for mild cognitive impairment and Alzheimer's disease prediction, outperforming the compared method by a large margin.

LinFeng Li, Jian Zhao, Zepeng Yang et al.

We present a winning solution to RoboSense 2025 Track 4: Cross-Modal Drone Navigation. The task retrieves the most relevant geo-referenced image from a large multi-platform corpus (satellite/drone/ground) given a natural-language query. Two obstacles are severe inter-platform heterogeneity and a domain gap between generic training descriptions and platform-specific test queries. We mitigate these with a domain-aligned preprocessing pipeline and a Mixture-of-Experts (MoE) framework: (i) platform-wise partitioning, satellite augmentation, and removal of orientation words; (ii) an LLM-based caption refinement pipeline to align textual semantics with the distinct visual characteristics of each platform. Using BGE-M3 (text) and EVA-CLIP (image), we train three platform experts using a progressive two-stage, hard-negative mining strategy to enhance discriminative power, and fuse their scores at inference. The system tops the official leaderboard, demonstrating robust cross-modal geo-localization under heterogeneous viewpoints.

Liang Lin, Zhihao Xu, Junhao Dong et al.

Large language model (LLM) alignment faces a critical dilemma when addressing multiple human preferences: improvements in one dimension frequently come at the expense of others, creating unavoidable trade-offs between competing objectives like helpfulness and harmlessness. While prior work mainly focuses on constraint-based optimization algorithms and data selection strategies to mitigate conflicts, these approaches overlook the fundamental issue of resolving conflicts directly at the parameter level. In this paper, we present OrthAlign, an innovative approach that pioneers a new paradigm by leveraging orthogonal subspace decomposition to fundamentally resolve gradient-level conflicts in multi-objective preference alignment. OrthAlign strategically decomposes parameter update spaces into orthogonal subspaces, ensuring that optimization toward different preferences occurs in mathematically non-interfering directions. Building upon this, we provide theoretical guarantees demonstrating that when parameter increments satisfy both orthogonal subspace constraints and spectral norm bounds, the resulting updates exhibit linear Lipschitz growth rather than exponential instability, ensuring stable convergence across all preference dimensions. Extensive experiments show that: I. OrthAlign achieves maximum single-preference improvements ranging from 34.61% to 50.89% after multiple-objective alignment across helpful, harmless, and truthful dimensions. II. With an average overall reward improvement of 13.96%.

Xiuyuan Chen, Jian Zhao, Yuchen Yuan et al.

Existing safety evaluation methods for large language models (LLMs) suffer from inherent limitations, including evaluator bias and detection failures arising from model homogeneity, which collectively undermine the robustness of risk evaluation processes. This paper seeks to re-examine the risk evaluation paradigm by introducing a theoretical framework that reconstructs the underlying risk concept space. Specifically, we decompose the latent risk concept space into three mutually exclusive subspaces: the explicit risk subspace (encompassing direct violations of safety guidelines), the implicit risk subspace (capturing potential malicious content that requires contextual reasoning for identification), and the non-risk subspace. Furthermore, we propose RADAR, a multi-agent collaborative evaluation framework that leverages multi-round debate mechanisms through four specialized complementary roles and employs dynamic update mechanisms to achieve self-evolution of risk concept distributions. This approach enables comprehensive coverage of both explicit and implicit risks while mitigating evaluator bias. To validate the effectiveness of our framework, we construct an evaluation dataset comprising 800 challenging cases. Extensive experiments on our challenging testset and public benchmarks demonstrate that RADAR significantly outperforms baseline evaluation methods across multiple dimensions, including accuracy, stability, and self-evaluation risk sensitivity. Notably, RADAR achieves a 28.87% improvement in risk identification accuracy compared to the strongest baseline evaluation method.

Yiyi Zhang, Yuchen Yuan, Ying Zheng et al.

Surgical triplet recognition, which involves identifying instrument, verb, target, and their combinations, is a complex surgical scene understanding challenge plagued by long-tailed data distribution. The mainstream multi-task learning paradigm benefiting from cross-task collaborative promotion has shown promising performance in identifying triples, but two key challenges remain: 1) inter-task optimization conflicts caused by entangling task-generic and task-specific representations; 2) intra-task optimization conflicts due to class-imbalanced training data. To overcome these difficulties, we propose the MLLM-Engaged Joint Optimization (MEJO) framework that empowers both inter- and intra-task optimization for surgical triplet recognition. For inter-task optimization, we introduce the Shared-Specific-Disentangled (S$^2$D) learning scheme that decomposes representations into task-shared and task-specific components. To enhance task-shared representations, we construct a Multimodal Large Language Model (MLLM) powered probabilistic prompt pool to dynamically augment visual features with expert-level semantic cues. Additionally, comprehensive task-specific cues are modeled via distinct task prompts covering the temporal-spatial dimensions, effectively mitigating inter-task ambiguities. To tackle intra-task optimization conflicts, we develop a Coordinated Gradient Learning (CGL) strategy, which dissects and rebalances the positive-negative gradients originating from head and tail classes for more coordinated learning behaviors. Extensive experiments on the CholecT45 and CholecT50 datasets demonstrate the superiority of our proposed framework, validating its effectiveness in handling optimization conflicts.

Zhenbo Xu, Wei Zhang, Xiao Tan et al.

Multiple-object tracking and segmentation (MOTS) is a novel computer vision task that aims to jointly perform multiple object tracking (MOT) and instance segmentation. In this work, we present PointTrack++, an effective on-line framework for MOTS, which remarkably extends our recently proposed PointTrack framework. To begin with, PointTrack adopts an efficient one-stage framework for instance segmentation, and learns instance embeddings by converting compact image representations to un-ordered 2D point cloud. Compared with PointTrack, our proposed PointTrack++ offers three major improvements. Firstly, in the instance segmentation stage, we adopt a semantic segmentation decoder trained with focal loss to improve the instance selection quality. Secondly, to further boost the segmentation performance, we propose a data augmentation strategy by copy-and-paste instances into training images. Finally, we introduce a better training strategy in the instance association stage to improve the distinguishability of learned instance embeddings. The resulting framework achieves the state-of-the-art performance on the 5th BMTT MOTChallenge.

Zhaoyi Yan, Yuchen Yuan, Wangmeng Zuo et al.

In this paper, we propose a novel perspective-guided convolution (PGC) for convolutional neural network (CNN) based crowd counting (i.e. PGCNet), which aims to overcome the dramatic intra-scene scale variations of people due to the perspective effect. While most state-of-the-arts adopt multi-scale or multi-column architectures to address such issue, they generally fail in modeling continuous scale variations since only discrete representative scales are considered. PGCNet, on the other hand, utilizes perspective information to guide the spatially variant smoothing of feature maps before feeding them to the successive convolutions. An effective perspective estimation branch is also introduced to PGCNet, which can be trained in either supervised setting or weakly-supervised setting when the branch has been pre-trained. Our PGCNet is single-column with moderate increase in computation, and extensive experimental results on four benchmark datasets show the improvements of our method against the state-of-the-arts. Additionally, we also introduce Crowd Surveillance, a large scale dataset for crowd counting that contains 13,000+ high-resolution images with challenging scenarios.

Zhuojun Chen, Junhao Cheng, Yuchen Yuan et al.

We seek to improve crowd counting as we perceive limits of currently prevalent density map estimation approach on both prediction accuracy and time efficiency. We leverage multilevel pixelation of density map as it helps improve SNR of training data and therefore, reduce prediction error. To achieve a better model, we introduce multilayer gradient fusion for training a density-aware global count regressor. More specifically, on training stage, a backbone network receives gradients from multiple branches to learn the density information, whereas those branches are to be detached to accelerate inference. By taking advantages of such method, our model improves benchmark results on public datasets and exhibits itself to be a new solution to crowd counting problems in practice.

Ming Sun, Yuchen Yuan, Feng Zhou et al.

Attention-based learning for fine-grained image recognition remains a challenging task, where most of the existing methods treat each object part in isolation, while neglecting the correlations among them. In addition, the multi-stage or multi-scale mechanisms involved make the existing methods less efficient and hard to be trained end-to-end. In this paper, we propose a novel attention-based convolutional neural network (CNN) which regulates multiple object parts among different input images. Our method first learns multiple attention region features of each input image through the one-squeeze multi-excitation (OSME) module, and then apply the multi-attention multi-class constraint (MAMC) in a metric learning framework. For each anchor feature, the MAMC functions by pulling same-attention same-class features closer, while pushing different-attention or different-class features away. Our method can be easily trained end-to-end, and is highly efficient which requires only one training stage. Moreover, we introduce Dogs-in-the-Wild, a comprehensive dog species dataset that surpasses similar existing datasets by category coverage, data volume and annotation quality. This dataset will be released upon acceptance to facilitate the research of fine-grained image recognition. Extensive experiments are conducted to show the substantial improvements of our method on four benchmark datasets.