论文检索

Xihong Yang, Siwei Wang, Jiaqi Jin, Fangdi Wang, Tianrui Liu, Yueming Jin, Xinwang Liu, En Zhu, Kunlun He

Multi-view clustering (MVC) aims to explore the common clustering structure across multiple views. Many existing MVC methods heavily rely on the assumption of view consistency, where alignments for corresponding samples across different views are ordered in advance. However, real-world scenarios often present a challenge as only partial data is consistently aligned across different views, restricting the overall clustering performance. In this work, we consider the model performance decreasing phenomenon caused by data order shift (i.e., from fully to partially aligned) as a generalized multi-view clustering problem. To tackle this problem, we design a causal multi-view clustering network, termed CauMVC. We adopt a causal modeling approach to understand multi-view clustering procedure. To be specific, we formulate the partially aligned data as an intervention and multi-view clustering with partially aligned data as an post-intervention inference. However, obtaining invariant features directly can be challenging. Thus, we design a Variational Auto-Encoder for causal learning by incorporating an encoder from existing information to estimate the invariant features. Moreover, a decoder is designed to perform the post-intervention inference. Lastly, we design a contrastive regularizer to capture sample correlations. To the best of our knowledge, this paper is the first work to deal generalized multi-view clustering via causal learning. Empirical experiments on both fully and partially aligned data illustrate the strong generalization and effectiveness of CauMVC.

Hao Si, Ehsan Javanmardi, Manabu Tsukada

Collaborative perception enables vehicles to overcome individual perception limitations by sharing information, allowing them to see further and through occlusions. In real-world scenarios, models on different vehicles are often heterogeneous due to manufacturer variations. Existing methods for heterogeneous collaborative perception address this challenge by fine-tuning adapters or the entire network to bridge the domain gap. However, these methods are impractical in real-world applications, as each new collaborator must undergo joint training with the ego vehicle on a dataset before inference, or the ego vehicle stores models for all potential collaborators in advance. Therefore, we pose a new question: Can we tackle this challenge directly during inference, eliminating the need for joint training? To answer this, we introduce Progressive Heterogeneous Collaborative Perception (PHCP), a novel framework that formulates the problem as few-shot unsupervised domain adaptation. Unlike previous work, PHCP dynamically aligns features by self-training an adapter during inference, eliminating the need for labeled data and joint training. Extensive experiments on the OPV2V dataset demonstrate that PHCP achieves strong performance across diverse heterogeneous scenarios. Notably, PHCP achieves performance comparable to SOTA methods trained on the entire dataset while using only a small amount of unlabeled data.

Yazhou Xing, Yang Fei, Yingqing He, Jingye Chen, Jiaxin Xie, Xiaowei Chi, Qifeng Chen

Learning a robust video Variational Autoencoder (VAE) is essential for reducing video redundancy and facilitating efficient video generation. Directly applying image VAEs to individual frames in isolation results in temporal inconsistencies and fails to compress temporal redundancy effectively. Existing works on Video VAEs compress temporal redundancy but struggle to handle videos with large motion effectively. They suffer from issues such as severe image blur and loss of detail in scenarios with large motion. In this paper, we present a powerful video VAE named VideoVAE+ that effectively reconstructs videos with large motion. First, we investigate two architecture choices and propose our simple yet effective architecture with better spatiotemporal joint modeling performance. Second, we propose to leverage the textual information in existing text-to-video datasets and incorporate text guidance during training. The textural guidance is optional during inference. We find that this design enhances the reconstruction quality and preservation of detail. Finally, our models achieve strong performance compared with various baseline approaches in both general videos and large motion videos, demonstrating its effectiveness on the challenging large motion scenarios.

Zixuan Hu, Dongxiao Li, Xinzhu Ma, Shixiang Tang, Xiaotong Li, Wenhan Yang, Ling-Yu Duan

Accurate monocular 3D object detection (M3OD) is pivotal for safety-critical applications like autonomous driving, yet its reliability deteriorates significantly under real-world domain shifts caused by environmental or sensor variations. To address these shifts, Test-Time Adaptation (TTA) methods have emerged, enabling models to adapt to target distributions during inference. While prior TTA approaches recognize the positive correlation between low uncertainty and high generalization ability, they fail to address the dual uncertainty inherent to M3OD: semantic uncertainty (ambiguous class predictions) and geometric uncertainty (unstable spatial localization). To bridge this gap, we propose Dual Uncertainty Optimization (DUO), the first TTA framework designed to jointly minimize both uncertainties for robust M3OD. Through a convex optimization lens, we introduce an innovative convex structure of the focal loss and further derive a novel conjugate loss, enabling label-agnostic uncertainty weighting and balanced learning for high-uncertainty objects. In parallel, we design a semantic-aware normal field constraint that preserves geometric coherence in regions with clear semantic cues, reducing uncertainty from the unstable 3D representation. This dual-branch mechanism forms a complementary loop: enhanced spatial perception improves semantic classification, and robust semantic predictions further refine spatial understanding. Extensive experiments demonstrate the superiority of DUO over existing methods across various datasets and domain shift types. The source code is available at https://github.com/hzcar/DUO.

Guanghui Shi, Xuefeng Liang, Wenjie Li, Xiaoyu Lin

Learning fine-grained representations from coarse labels for fine-grained visual recognition (FGVR) is a challenging yet valuable task, as it alleviates the reliance on labor-intensive fine-grained annotations. Early approaches focused primarily on minimizing intra-fine-grained-class variation but overlooked inter-fine-grained-class separability, resulting in limited FGVR performance. Subsequent studies employed a top-down paradigm to enhance separability via deep clustering, yet these methods require predefining the number of fine-grained classes, which is often impractical to obtain. Here, we introduce a bottom-up learning paradigm that constructs a hierarchical dendrogram by iteratively merging similar instances/clusters, inferring higher-level semantics from lowest-level instances without predefining class numbers. Leveraging this, we propose BuCSFR, a novel method that integrates a Bottom-up Construction (BuC) module to build the dendrogram based on a minimal information loss criterion, and a Separable Fine-grained Representation (SFR) module that treats dendrogram nodes as pseudo-labels to ensure representation separability. The synergistic interaction between these modules enables iterative enhancement, grounded theoretically in the Expectation-Maximization (EM) framework. Extensive experiments on five benchmark datasets demonstrate the superiority of our approach, showcasing its effectiveness in learning separable representations for FGVR. The source code is available at: https://github.com/BeCarefulOfYournaoke/BuCSFR.

Akshat Ramachandran, Mingyu Lee, Huan Xu, Souvik Kundu, Tushar Krishna

We present OuroMamba, the first data-free post-training quantization (DFQ) method for vision Mamba-based models (VMMs). We identify two key challenges in enabling DFQ for VMMs, (1) VMM's recurrent state transitions restricts the capturing of long-range interactions and leads to semantically weak synthetic data, (2) VMM activations exhibit dynamic outlier variations across time-steps, rendering existing static PTQ techniques ineffective. To address these challenges, OuroMamba presents a two-stage framework: (1) OuroMamba-Gen to generate semantically rich and meaningful synthetic data. It applies contrastive learning on patch level VMM features generated through neighborhood interactions in the latent state space, (2) OuroMamba-Quant to employ mixed-precision quantization with lightweight dynamic outlier detection during inference. In specific, we present a thresholding based outlier channel selection strategy for activations that gets updated every time-step. Extensive experiments across vision and generative tasks show that our data-free OuroMamba surpasses existing data-driven PTQ techniques, achieving state-of-the-art performance across diverse quantization settings. Additionally, we implement efficient GPU kernels to achieve practical latency speedup of up to 2.36x. Code and synthetic dataset are available here: https://github.com/georgia-tech-synergy-lab/ICCV-OuroMamba.

Junho Kim, Hyungjin Chung, Byung-Hoon Kim

Category-agnostic pose estimation (CAPE) has traditionally relied on support images with annotated keypoints, a process that is often cumbersome and may fail to fully capture the necessary correspondences across diverse object categories. Recent efforts have explored the use of text queries, leveraging their enhanced stability and generalization capabilities. However, existing approaches often remain constrained by their reliance on support queries, their failure to fully utilize the rich priors embedded in pre-trained large language models, and the limitations imposed by their parametric distribution assumptions. To address these challenges, we introduce CapeLLM, the first multimodal large language model (MLLM) designed for CAPE. Our method only employs query image and detailed text descriptions as an input to estimate category-agnostic keypoints. Our method encompasses effective training strategies and carefully designed instructions for applying the MLLM to CAPE. Moreover, we propose an inference mechanism that further enhances the reasoning process for unseen keypoints. while flexibly modeling their underlying spatial distribution and uncertainty, allowing for adaptive refinement based on contextual cues. We conducted extensive experiments to apply the MLLM to CAPE effectively, focusing not only on the model architecture and prompt design but also on ensuring robustness across input variations. Our approach sets a new state-of-the-art on the MP-100 benchmark in the 1-shot and even 5-shot setting, marking a significant advancement in the field of category-agnostic pose estimation. Code is available https://github.com/Junhojuno/CapeLLM

Yu Cheng, Fajie Yuan

Recent advances in Latent Video Diffusion Models (LVDMs) have revolutionized video generation by leveraging Video Variational Autoencoders (Video VAEs) to compress intricate video data into a compact latent space. However, as LVDM training scales, the computational overhead of Video VAEs becomes a critical bottleneck, particularly for encoding high-resolution videos. To address this, we propose LeanVAE, a novel and ultra-efficient Video VAE framework that introduce two key innovations: (1) a lightweight architecture based on a Neighborhood-Aware Feedforward (NAF) module and non-overlapping patch operations, drastically reducing computational cost, and (2) the integration of wavelet transforms and compressed sensing techniques to enhance reconstruction quality. Extensive experiments validate LeanVAE's superiority in video reconstruction and generation, particularly in enhancing efficiency over existing Video VAEs. Our model offers up to 50x fewer FLOPs and 44x faster inference speed while maintaining competitive reconstruction quality, providing insights for scalable, efficient video generation. Our models and code are available at https://github.com/westlake-repl/LeanVAE.

Quankai Gao, Iliyan Georgiev, Tuanfeng Y. Wang, Krishna Kumar Singh, Ulrich Neumann, Jae Shin Yoon

3D generation has made significant progress, however, it still largely remains at the object-level. Feedforward 3D scene-level generation has been rarely explored due to the lack of models capable of scaling-up latent representation learning on 3D scene-level data. Unlike object-level generative models, which are trained on well-labeled 3D data in a bounded canonical space, scene-level generations with 3D scenes represented by 3D Gaussian Splatting (3DGS) are unbounded and exhibit scale inconsistency across different scenes, making unified latent representation learning for generative purposes extremely challenging. In this paper, we introduce Can3Tok, the first 3D scene-level variational autoencoder (VAE) capable of encoding a large number of Gaussian primitives into a low-dimensional latent embedding, which effectively captures both semantic and spatial information of the inputs. Beyond model design, we propose a general pipeline for 3D scene data processing to address scale inconsistency issue. We validate our method on the recent scene-level 3D dataset DL3DV-10K, where we found that only Can3Tok successfully generalizes to novel 3D scenes, while compared methods fail to converge on even a few hundred scene inputs during training and exhibit zero generalization ability during inference. Finally, we demonstrate image-to-3DGS and text-to-3DGS generation as our applications to demonstrate it's ability to faciliate downstream generation tasks. Code will be released.

Sihang Li, Siqi Tan, Bowen Chang, Jing Zhang, Chen Feng, Yiming Li

Visual localization, which estimates a camera's pose within a known scene, is a fundamental capability for autonomous systems. While absolute pose regression (APR) methods have shown promise for efficient inference, they often struggle with generalization. Recent approaches attempt to address this through data augmentation with varied viewpoints, yet they overlook a critical factor: appearance diversity.In this work, we identify appearance variation as the key to robust localization. Specifically, we first lift real 2D images into 3D Gaussian Splats with varying appearance and deblurring capabilities, enabling the synthesis of diverse training data that varies not just in poses but also in environmental conditions such as lighting and weather. To fully unleash the potential of the appearance-diverse data, we build a two-branch joint training pipeline with an adversarial discriminator to bridge the syn-to-real gap.Extensive experiments demonstrate that our approach significantly outperforms state-of-the-art methods, reducing translation and rotation errors by 50% and 22% on indoor datasets, and 37% and 42% on outdoor datasets. Most notably, our method shows remarkable robustness in dynamic driving scenarios under varying weather conditions and in day-to-night scenarios, where previous APR methods fail.

Zengyu Wan, Wei Zhai, Yang Cao, Zhengjun Zha

Visual 3D motion estimation aims to infer the motion of 2D pixels in 3D space based on visual cues. The key challenge arises from depth variation induced spatio-temporal motion inconsistencies, disrupting the assumptions of local spatial or temporal motion smoothness in previous motion estimation frameworks. In contrast, event cameras offer new possibilities for 3D motion estimation through continuous adaptive pixel-level responses to scene changes. This paper presents EMoTive, a novel event-based framework that models spatio-temporal trajectories via event-guided non-uniform parametric curves, effectively characterizing locally heterogeneous spatio-temporal motion. Specifically, we first introduce Event Kymograph - an event projection method that leverages a continuous temporal projection kernel and decouples spatial observations to encode fine-grained temporal evolution explicitly. For motion representation, we introduce a density-aware adaptation mechanism to fuse spatial and temporal features under event guidance, coupled with a non-uniform rational curve parameterization framework to adaptively model heterogeneous trajectories. The final 3D motion estimation is achieved through multi-temporal sampling of parametric trajectories, yielding optical flow and depth motion fields. To facilitate evaluation, we introduce CarlaEvent3D, a multi-dynamic synthetic dataset for comprehensive validation. Extensive experiments on both this dataset and a real-world benchmark demonstrate the effectiveness of the proposed method.

Kaixiang Yang, Xin Li, Qiang Li, Zhiwei Wang

Anticipating and recognizing surgical workflows are critical for intelligent surgical assistance systems. However, existing methods rely on deterministic decision-making, struggling to generalize across the large anatomical and procedural variations inherent in real-world surgeries. In this paper, we introduce an innovative framework that incorporates stochastic modeling through a denoising diffusion probabilistic model (DDPM) into conventional deterministic learning for surgical workflow analysis. At the heart of our approach is a collaborative co-training paradigm: the DDPM branch captures procedural uncertainties to enrich feature representations, while the task branch focuses on predicting surgical phases and instrument usage. Theoretically, we demonstrate that this mutual refinement mechanism benefits both branches: the DDPM reduces prediction errors in uncertain scenarios, and the task branch directs the DDPM toward clinically meaningful representations. Notably, the DDPM branch is discarded during inference, enabling real-time predictions without sacrificing accuracy. Experiments on the Cholec80 dataset show that for the anticipation task, our method achieves a 16% reduction in eMAE compared to state-of-the-art approaches, and for phase recognition, it improves the Jaccard score by 1.0%. Additionally, on the AutoLaparo dataset, our method achieves a 1.5% improvement in the Jaccard score for phase recognition, while also exhibiting robust generalization to patient-specific variations. Our code and weight will be available.

Andong Zhu 0001, Sheng Zhang 0001, Xiaohang Shi 0001, Hesheng Sun, Yu Liang 0001, Zhuzhong Qian, Han Zheng, Xiaokun Wang 0002, Ning Jiang

Video analytics pipelines migrating to edge deployments are facing performance bottlenecks under limited bandwidth. Non-uniform intra-frame encoding emerges to further compress pixels without affecting the output of the server deep neural network (DNN), while it is inefficient in high-resolution video streaming at low bandwidth. The detail enhancement capability of neural super-resolution (SR) permits resolution downsampling and aggressive compression on edge devices for low-latency transmission. To exploit its accuracy potential, DNN-oriented non-uniform encoding is expected to be additionally aware of SR models. However, traditional codecs struggle to cope with both quality optimization for SR and global semantic features for DNN. We advocate neural codecs for coordinated encoding and enhancement, enabling analytic-oriented video streaming with optimal accuracy-delay tradeoffs. Our system, VidIQ, achieves quality-enhanced real-time video analytics by 1) improving the network architecture of neural codecs (at two granularity) to integrate SR models into a DNN-oriented analytics pipeline, and 2) adapting the multi-scale encoder and SR-decoder to scene dynamics (i.e., content and bandwidth variations) with the help of the monolithic controller to hold a performance advantage. Extensive evaluations showcase that VidIQ reduces end-to-end delay by 35.8% and improves analytics accuracy by 21.2% compared to the recent video compression, enhancement, and streaming baselines.

Jiazhen Chen, Zheng Ma 0011, Sichao Fu, Mingbin Feng, Tony S. Wirjanto, Weihua Ou

Graphs play a pivotal role in multimedia applications by integrating information to model complex relationships. Recently, graph class-incremental learning (GCIL) has garnered attention, allowing graph neural networks (GNNs) to adapt to evolving graph analytical tasks by incrementally learning new class knowledge while retaining knowledge of old classes. Existing GCIL methods primarily focus on a closed-set assumption, where all test samples are presumed to belong to previously known classes. Such assumption restricts their applicability in real-world scenarios, where unknown classes naturally emerge during inference, and are absent during training. In this paper, we explore a more challenging open-set graph class-incremental learning scenario with two intertwined challenges: catastrophic forgetting of old classes, which impairs the detection of unknown classes, and inadequate open-set recognition, which destabilizes the retention of learned knowledge. To address the above problems, a novel OGCIL framework is proposed, which utilizes pseudo-sample embedding generation to effectively mitigate catastrophic forgetting and enable robust detection of unknown classes. To be specific, a prototypical conditional variational autoencoder is designed to synthesize node embeddings for old classes, enabling knowledge replay without storing raw graph data. To handle unknown classes, we employ a mixing-based strategy to generate out-of-distribution (OOD) samples from pseudo in-distribution and current node embeddings. A novel prototypical hypersphere classification loss is further proposed, which anchors in-distribution embeddings to their respective class prototypes, while repelling OOD embeddings away. Instead of assigning all unknown samples into one cluster, our proposed objective function explicitly models them as outliers through prototype-aware rejection regions, ensuring a robust open-set recognition. Extensive experiments on five benchmarks demonstrate the effectiveness of OGCIL over existing GCIL and open-set GNN methods.

Hongjun Liu, Chao Yao, Yalan Zhang, Xiaokun Wang 0001, Xiaojuan Ban

Electroencephalogram (EEG) signal classification faces significant challenges due to data distribution shifts caused by heterogeneous electrode configurations, acquisition protocols, and hardware discrepancies across domains. This paper introduces IMAC, a novel channel-dependent mask and imputation self-supervised framework that formulates the alignment of cross-domain EEG data shifts as a spatial time series imputation task. To address heterogeneous electrode configurations in cross-domain scenarios, IMAC first standardizes different electrode layouts using a 3D-to-2D positional unification mapping strategy, establishing unified spatial representations. Unlike previous mask-based self-supervised representation learning methods, IMAC introduces spatio-temporal signal alignment. This involves constructing a channel-dependent mask and reconstruction task framed as a low-to-high resolution EEG spatial imputation problem. Consequently, this approach simulates cross-domain variations such as channel omissions and temporal instabilities, thus enabling the model to leverage the proposed imputer for robust signal alignment during inference. Furthermore, IMAC incorporates a disentangled structure that separately models the temporal and spatial information of the EEG signals separately, reducing computational complexity while enhancing flexibility and adaptability. Comprehensive evaluations across 10 publicly available EEG datasets demonstrate IMAC's superior performance, achieving state-of-the-art classification accuracy in both cross-subject and cross-center validation scenarios. Notably, IMAC shows strong robustness under both simulated and real-world distribution shifts, surpassing baseline methods by up to 35% in integrity scores while maintaining consistent classification accuracy.

Changzhou Li, Xinyu Yang 0001, Weiguo Yang, Xinyi Li

Efficient and precise open-vocabulary 3D scene segmentation remains a critical challenge in computer vision. While current leading methods encode CLIP language features into 3D Gaussians to achieve high segmentation accuracy and fast inference speeds, they suffer from point ambiguity issues caused by separately training on multi-level 2D semantic masks. This approach not only compromises time and space efficiency but also degrades accuracy when selecting optimal semantic levels. To overcome these limitations, we propose Voxel-Aware Fusion Language Gaussian Splatting (VaF-LangSplat), a novel framework that jointly optimizes geometric and semantic representations. Our approach first voxelizes 3D Gaussians using sparse point clouds and lightweight MLP decoders, effectively disentangling language features from geometric attributes. This enables simultaneous training across arbitrary semantic levels with minimal overhead. Crucially, we introduce Fusion Language Splatting, which aligns geometric and multi-level semantic distributions to sharpen boundary definitions while eliminating redundant Gaussian expansions. The voxel-aware representation further enhances robustness against motion blur and lighting variations. Experiments on open-vocabulary 3D localization and segmentation tasks demonstrate that VaF-LangSplat outperforms LangSplat (the prior state-of-the-art) with significant improvements in both segmentation/localization accuracy and efficiency: 4X faster training and 15X reduced storage requirements.

Yiwen Liang, Hui Chen 0013, Yizhe Xiong, Zihan Zhou, Mengyao Lyu, Zijia Lin, Shuaicheng Niu, Sicheng Zhao, Jungong Han, Guiguang Ding

Vision-language models (VLMs) exhibit remarkable zero-shot capabilities but struggle with distribution shifts in downstream tasks when labeled data is unavailable, which has motivated the development of Test-Time Adaptation (TTA) to improve VLMs' performance during inference without annotations. Among various TTA approaches, cache-based methods show promise by preserving historical knowledge from low-entropy samples in a dynamic cache and fostering efficient adaptation. However, these methods face two critical reliability challenges: (1) entropy often becomes unreliable under distribution shifts, causing error accumulation in the cache and degradation in adaptation performance; (2) the final predictions may be unreliable due to inflexible decision boundaries that fail to accommodate large downstream shifts. To address these challenges, we propose a Reliable Test-time Adaptation (ReTA) method that integrates two complementary strategies to enhance reliability from two perspectives. First, to mitigate the unreliability of entropy as a sample selection criterion for cache construction, we introduce Consistency-aware Entropy Reweighting (CER), which incorporates consistency constraints to weight entropy during cache updating. While conventional approaches rely solely on low entropy for cache prioritization and risk introducing noise, our method leverages predictive consistency to maintain a high-quality cache and facilitate more robust adaptation. Second, we present Diversity-driven Distribution Calibration (DDC), which models class-wise text embeddings as multivariate Gaussian distributions, enabling adaptive decision boundaries for more accurate predictions across visually diverse content. Extensive experiments demonstrate that ReTA consistently outperforms state-of-the-art methods, particularly under real-world distribution shifts.

Dexuan Xu, Yanyuan Chen, Yu Huang 0004, Shihao E, Yiwei Lou, Yongzhi Cao, Hanpin Wang, Meikang Qiu

Medical Visual Question Answering (Medical VQA) plays an important role in medical informatics. However, the robustness of existing medical VQA models is severely challenged by adversarial attacks. Current methods (e.g. adversarial training and noise-based reasoning) heavily rely on additional data or complex procedures and often ignore model-level robustness. To address these issues, we propose Multimodal Variational Masked Autoencoder (MVMAE), a novel pre-training framework designed to enhance the robustness of the medical VQA task. MVMAE leverages masked modeling and variational inference to extract robust multimodal features. The framework introduces a low-cost multimodal bottleneck fusion module and employs reparameterization to sample robust latent representations, ensuring effective feature fusion and reconstruction. Extensive experiments on public medical VQA datasets demonstrate that MVMAE significantly improves resistance to various adversarial attacks and outperforms other medical multimodal pre-training methods.

Jingxing Guo, Guilian Chen, Yimu Sun, Huisi Wu, Jing Qin 0001

Automatic segmentation of echocardiography videos is crucial for computer-aided cardiovascular function assessment in clinical practice. However, it is a challenging task owing to the existence of massive speckle noise, the large shape variations of heart structures between frames, and limited annotations. In this paper, we propose a novel semi-supervised video segmentation model to comprehensively meet these challenges. The proposed approach has two key techniques. First, we propose a dual-stream architecture that processes spatial and temporal features through separate pathways to capture structural details and motion patterns, then enhances spatiotemporal representations by interacting these decomposed features with query features generated from the original input. Second, as speckle noise primarily concentrates in high-frequency regions, we extend the traditional dilated convolution from a frequency perspective, enabling it to adaptively adjust the dilation rate and convolution kernel weights based on high frequency speckle noise information. This enables the network to focus on specific frequency bands, thereby enhancing its ability to capture both low-frequency context and high-frequency local details. Extensive experiments on the CAMUS and EchoNet-Dynamic datasets demonstrate that our method outperforms existing state-of-the-art methods in terms of both accuracy and inference speed. Codes are available at https://github.com/guojx2255/HSCA-SDC.

Yudong Zhang 0008, Ruobing Xie, Xingwu Sun, Yiqing Huang, Jiansheng Chen 0001, Zhanhui Kang, Di Wang 0052, Yu Wang 0002

Large vision-language models (LVLMs) have demonstrated exceptional performance on complex multimodal tasks. However, they continue to suffer from significant hallucination issues, including object, attribute, and relational hallucinations. To accurately detect these hallucinations, we investigated the variations in cross-modal attention patterns between hallucination and non-hallucination states. Leveraging these distinctions, we developed a lightweight detector capable of identifying hallucinations. Our proposed method, Detecting Hallucinations by Cross-modal Attention Patterns (DHCP), is straightforward and does not require additional LVLM training or extra LVLM inference steps. Experimental results show that DHCP achieves remarkable performance in hallucination detection. By offering novel insights into the identification and analysis of hallucinations in LVLMs, DHCP contributes to advancing the reliability and trustworthiness of these models. The code is available at https://github.com/btzyd/DHCP.