论文检索

Jinxing Zhou, Yanghao Zhou, Mingfei Han, Tong Wang, Xiaojun Chang, Hisham Cholakkal, Rao Muhammad Anwer

Referring Audio-Visual Segmentation (Ref-AVS) aims to segment target objects in audible videos based on given reference expressions. Prior works typically rely on learning latent embeddings via multimodal fusion to prompt a tunable SAM/SAM2 decoder for segmentation, which requires strong pixel-level supervision and lacks interpretability. From a novel perspective of explicit reference understanding, we propose TGS-Agent, which decomposes the task into a Think-Ground-Segment process, mimicking the human reasoning procedure by first identifying the referred object through multimodal analysis, followed by coarse-grained grounding and precise segmentation. To this end, we first propose Ref-Thinker, a multimodal language model capable of reasoning over textual, visual, and auditory cues. We construct an instruction-tuning dataset with explicit object-aware think-answer chains for Ref-Thinker fine-tuning. The object description inferred by Ref-Thinker is used as an explicit prompt for Grounding-DINO and SAM2, which perform grounding and segmentation without relying on pixel-level supervision. Additionally, we introduce R2-AVSBench, a new benchmark with linguistically diverse and reasoning-intensive references for better evaluating model generalization. Our approach achieves state-of-the-art results on both standard Ref-AVSBench and proposed R2-AVSBench.

Delong Zhao, Qiang Huang, Di Yan, Yiqun Sun, Jun Yu

Concept Bottleneck Models (CBMs) enhance interpretability by introducing a layer of human-understandable concepts between inputs and predictions. While recent methods automate concept generation using Large Language Models (LLMs) and Vision-Language Models (VLMs), they still face three fundamental challenges: poor visual grounding, concept redundancy, and the absence of principled metrics to balance predictive accuracy and concept compactness. We introduce PS-CBM, a Partially Shared CBM framework that addresses these limitations through three core components: (1) a multimodal concept generator that integrates LLM-derived semantics with exemplar-based visual cues; (2) a Partially Shared Concept Strategy that merges concepts based on activation patterns to balance specificity and compactness; and (3) Concept-Efficient Accuracy (CEA), a post-hoc metric that jointly captures both predictive accuracy and concept compactness. Extensive experiments on eleven diverse datasets show that PS-CBM consistently outperforms state-of-the-art CBMs, improving classification accuracy by 1.0%–7.4% and CEA by 2.0%–9.5%, while requiring significantly fewer concepts. These results underscore PS-CBM’s effectiveness in achieving both high accuracy and strong interpretability.

CaiJie Zhao, Bob Zhang

Aiming to estimate the full extent of partially occluded objects, amodal segmentation is a critical capability for visual intelligence. Existing methods suffer from limitations in efficiency and precision, due to their reliance on auxiliary information or two-stage architectures. Furthermore, they lack generalizability, failing to meet practical requirements. To overcome these challenges, we proposed a new paradigm, CondDiff-AMO, that interprets amodal segmentation as a denoising problem by leveraging diffusion models. Methodologically, the designed novel framework consists of three key innovations to adapt the task characteristics and unlocks the diffusion models’ potential in amodal segmentation, including a masking strategy in the forward process, an adaptive transformer for conditional feature extraction, and visual-guided sampling. In the forward process, progressive masking strategy converts ground-truth masks to visible masks, simulating amodal segmentation process to enhance reasoning regarding occluded areas. For architectural design, a pyramid network with feature refinement extracts adaptive and representative conditional priors, improving the guidance in the denoising process of diffusion models. As for the sampling stage, a visible mask is incorporated with an ensemble strategy, restricting the prediction on occluded part. Experiments were conducted on five well-known datasets under supervised and zero-shot learning, with the results confirming that CondDiff-AMO outperforms state-of-the-art methods.

Hong Zhang, Yixuan Lyu, Hanyang Liu, Jianbo Song, Ding Yuan, Yifan Yang

Video Camouflaged Object Detection (VCOD) poses significant challenges due to the subtle appearance of camouflaged objects, especially under dynamic motion and occlusion. Existing methods predominantly rely on optical flow or black-box features for motion modeling, which often entail substantial computational costs and suffer from limited interpretability. Inspired by the human strategy of identifying abnormal movements between frames and the principle of event camera image formation, we propose an eventstream-inspired dual-branch framework for VCOD. Specifically, we design an eventstream-like data extraction module to capture pixel-level motion variations, effectively distinguishing object motion from background dynamics. This event-based representation is integrated into SAM2 through a dual-branch memory-augmented framework, consisting of Time Bridge Attention and Visual Bridge Attention, enabling joint modeling of motion and appearance cues. In addition, we introduce a Prompt Embedding Generator to eliminate the need for human-provided interactive prompts, facilitating fully automatic VCOD. Extensive experiments on MoCA-Mask and CAD2016 demonstrate that our approach significantly outperforms state-of-the-art methods, achieving both superior segmentation accuracy and interpretable motion modeling. To the best of our knowledge, this is the first work to incorporate eventstream-inspired representations into the VCOD task.

Guanghao Zhang, Tao Zhong, Yan Xia, Mushui Liu, Zhelun Yu, Haoyuan Li, Wanggui He, Dong She, Yi Wang, Hao Jiang

While previous multimodal slow-thinking methods have demonstrated remarkable success in single-image understanding scenarios, their effectiveness becomes fundamentally constrained when extended to more complex multi-image comprehension tasks. This limitation stems from their predominant reliance on text-based intermediate reasoning processes. While for human, when engaging in sophisticated multi-image analysis, they typically perform two complementary cognitive operations: (1) continuous cross-image visual comparison through region-of-interest matching, and (2) dynamic memorization of critical visual concepts throughout the reasoning chain. Motivated by these observations, we propose the Complex Multi-Modal Chain-of-Thought (CMMCoT) framework, a multi-step reasoning framework that mimics human-like "slow thinking" for multi-image understanding. Our approach incorporates two key innovations: (1) The construction of interleaved multimodal multi-step reasoning chains, which utilize critical visual region tokens, extracted from intermediate reasoning steps, as supervisory signals. This mechanism not only facilitates comprehensive cross-modal understanding but also enhances model interpretability. (2) The introduction of a test-time memory augmentation module that expands the model’s reasoning capacity during inference while preserving parameter efficiency. Furthermore, to facilitate research in this direction, we have curated a novel multi-image slow-thinking dataset. Extensive experiments demonstrate the effectiveness of our model.

Songyuan Yang, Wanrong Huang, Yinuo Liu, Zhang Ke-Di, Xihuai He, Shaowu Yang, Huibin Tan

Generating realistic and coordinated 3D human motion for multiple individuals within complex environments remains a significant challenge. Existing text-to-motion methods are often ``blind'' to the physical scene, leading to implausible motions, while scene-conditioned (HSI) approaches demand cumbersome full 3D data and largely neglect multi-person dynamics. To address these limitations, we introduce the VL2Motion paradigm and its embodiment, MMG-VL, a hierarchical framework that generates coordinated multi-person motions from the most accessible inputs: a single 2D image and natural language. MMG-VL first employs a Scene-Aware Intent Planner (SAIP) to interpret the visual context and decompose the user's command into a set of spatially-grounded, multi-person action blueprints. Subsequently, a Coordinated Motion Synthesizer (CMS) translates these blueprints into high-fidelity 3D motion sequences. The synergy between these stages is driven by two novel loss functions: a Spatial-Semantic Grounding Loss to ensure the planner's output is grounded in visual reality, and a Coordinated Environmental Realism Loss that enforces physical constraints and coherent group dynamics during synthesis. To facilitate this research, we introduce HumanVL, the first large-scale dataset featuring multi-person activities in multi-room scenes, providing aligned images, text, blueprints, 3D motions, and scene geometry. Extensive experiments demonstrate that MMG-VL significantly outperforms existing methods in generating spatially coherent, physically realistic, and coordinated multi-person motions, paving the way for more scalable and intuitive creation of dynamic virtual worlds.

Zhonghao Yan, Muxi Diao, Yuxuan Yang, Ruoyan Jing, Jiayuan Xu, Kaizhou Zhang, Lele Yang, Yanxi Liu, Kongming Liang, Zhanyu Ma

Accurately grounding regions of interest (ROIs) is critical for diagnosis and treatment planning in medical imaging. While multimodal large language models (MLLMs) combine visual perception with natural language, current medical-grounding pipelines still rely on supervised fine-tuning with explicit spatial hints, making them ill-equipped to handle the implicit queries common in clinical practice. This work makes three core contributions. We first define Unified Medical Reasoning Grounding (UMRG), a novel vision–language task that demands clinical reasoning and pixel-level grounding. Second, we release U-MRG-14K, a dataset of 14K samples featuring pixel-level masks alongside implicit clinical queries and reasoning traces, spanning 10 modalities, 15 super-categories, and 108 specific categories. Finally, we introduce MedReasoner, a modular framework that distinctly separates reasoning from segmentation: an MLLM reasoner is optimized with reinforcement learning, while a frozen segmentation expert converts spatial prompts into masks, with alignment achieved through format and accuracy rewards. MedReasoner achieves state-of-the-art performance on U-MRG-14K and demonstrates strong generalization to unseen clinical queries, underscoring the significant promise of reinforcement learning for interpretable medical grounding.

Shuhan Xu, Siyuan Liang, Hongling Zheng, Aishan Liu, Xinbiao Wang, Yong Luo, Fu Lin, Leszek Rutkowski, Dacheng Tao

Visual language models (VLMs) have made significant progress in image captioning tasks, yet recent studies have found they are vulnerable to backdoor attacks. Attackers can inject undetectable perturbations into the data during inference, triggering abnormal behavior and generating malicious captions. These attacks are particularly challenging to detect and defend against due to the stealthiness and cross-modal propagation of the trigger signals. In this paper, we identify two key vulnerabilities by analyzing existing attack patterns: (1) the model exhibits abnormal attention concentration on certain regions of the input image, and (2) backdoor attacks often induce semantic drift and sentence incoherence. Based on these insights, we propose Semantic Reward Defense (SRD), a reinforcement learning framework that mitigates backdoor behavior without requiring any prior knowledge of trigger patterns. SRD learns to apply discrete perturbations to sensitive contextual regions of image inputs via a deep Q-network policy, aiming to confuse attention and disrupt the activation of malicious paths. To guide policy optimization, we design a reward signal named semantic fidelity score, which jointly assesses the semantic consistency and linguistic fluency of the generated captions, encouraging the agent to achieve a robust yet faithful output. SRD offers a trigger-agnostic, policy-interpretable defense paradigm that effectively mitigates local (TrojVLM) and global (Shadowcast) backdoor attacks, reducing ASR to 3.6% and 5.6% respectively, with less than 15% average CIDEr drop on the clean inputs.

Mingjie Xu, Jinpeng Chen, Yuzhi Zhao, Jason Chun Lok Li, Yue Qiu, Zekang Du, Mengyang Wu, Pingping Zhang, Kun Li, Hongzheng Yang 等

Multimodal Large Language Models (MLLM) have enabled a wide range of advanced vision-language applications, including fine-grained object recognition and contextual understanding. When querying specific regions or objects in an image, human users naturally use "Visual Prompts" (VP) like bounding boxes to provide reference. However, no existing benchmark systematically evaluates the ability of MLLMs to interpret such VPs. This gap raises uncertainty about whether current MLLMs can effectively recognize VPs, an intuitive prompting method for humans, and utilize them to solve problems. To address this limitation, we introduce VP-Bench, aiming to assess MLLMs’ capability in VP perception and utilization. VP-Bench employs a two-stage evaluation framework: Stage 1 examines models’ ability to perceive VPs in natural scenes, utilizing 100K visualized prompts spanning 8 shapes and 355 attribute combinations. Stage 2 investigates the impact of VPs on downstream tasks, measuring their effectiveness in real-world problem-solving scenarios. Using VP-Bench, we evaluate 21 MLLMs, including proprietary systems (e.g., GPT-4o) and open-source models (e.g., InternVL-2.5 and Qwen2.5-VL). In addition, we conduct a comprehensive analysis of the factors influencing VP understanding, such as attribute variations and model scale. VP-Bench establishes a new reference framework for studying MLLMs’ ability to comprehend and resolve grounded referring questions.

Jiazheng Xu, Yu Huang, Jiale Cheng, Yuanming Yang, Jiajun Xu, Yuan Wang, Wenbo Duan, Shen Yang, Qunlin Jin, Shurun Li 等

Visual generative models have achieved remarkable progress in synthesizing photorealistic images and videos, yet aligning their outputs with human preferences across critical dimensions remains a persistent challenge. Though reinforcement learning from human feedback offers promise for preference alignment, existing reward models for visual generation face limitations, including black-box scoring without interpretability and potentially resultant unexpected biases. We present VisionReward, a general framework for learning human visual preferences in both image and video generation. Specifically, we employ a hierarchical visual assessment framework to capture fine-grained human preferences, and leverages linear weighting to enable interpretable preference learning. Furthermore, we propose a multi-dimensional consistent strategy when using VisionReward as a reward model during preference optimization for visual generation. Experiments show that VisionReward can significantly outperform existing image and video reward models on both machine metrics and human evaluation. Notably, VisionReward surpasses VideoScore by 17.2% in preference prediction accuracy, and text-to-video models with VisionReward achieve a 31.6% higher pairwise win rate compared to the same models using VideoScore.

Hongyang Wang, Yichen Shi, Zhuofu Tao, Yuhao Gao, Liepiao Zhang, Xun Lin, Jun Feng, Xiaochen Yuan, Zitong Yu, Xiaochun Cao

Face anti-spoofing (FAS) is crucial for protecting facial recognition systems from presentation attacks. Previous methods approached this task as a classification problem, lacking interpretability and reasoning behind the predicted results. Recently, multimodal large language models (MLLMs) have shown strong capabilities in perception, reasoning, and decision-making in visual tasks. However, there is currently no universal and comprehensive MLLM and dataset specifically designed for FAS task. To address this gap, we propose FaceShield, a MLLM for FAS, along with the corresponding pre-training and supervised fine-tuning (SFT) datasets, FaceShield-pre10K and FaceShield-sft45K. FaceShield is capable of determining the authenticity of faces, identifying types of spoofing attacks, providing reasoning for its judgments, and detecting attack areas. Specifically, we employ spoof-aware vision perception (SAVP) that incorporates both the original image and auxiliary information based on prior knowledge. We then use an prompt-guided vision token masking (PVTM) strategy to random mask vision tokens, thereby improving the model's generalization ability. We conducted extensive experiments on three benchmark datasets, demonstrating that FaceShield significantly outperforms previous deep learning models and general MLLMs on four FAS tasks, i.e., coarse-grained classification, fine-grained classification, reasoning, and attack localization.

Changshuo Wang, Zhijian Hu, Xiang Fang, Zai Yang Yu, Yibin Wu, Mingkun Xu, Yusong Wang, Xingyu Gao, Prayag Tiwari

Few-shot and zero-shot point cloud semantic segmentation aim to accurately segment novel categories using limited or no labeled samples, respectively. However, existing methods face significant challenges including domain shifts between support and query sets and the inability to handle both few-shot and zero-shot scenarios within a unified framework. To address these issues, we propose a biologically-inspired Evolutionary Domain Symbiosis Network EDS-Net for unified few-shot and zero-shot point cloud semantic segmentation. Specifically, inspired by natural symbiotic evolution, we propose a Symbiotic Evolution Module (SEM) that models co-adaptation between support and query features through self-correlation and cross-correlation mechanisms. Second, motivated by genetic crossover mechanisms, we introduce a Vision-Semantic Bridging Module (VSBM) that treats visual prototypes and semantic prototypes as two “parent” individuals, creating fused offspring prototypes through adaptive crossover operations and mutation strategies for zero-shot scenarios. Third, we develop a multi-generational evolutionary optimization framework employing an adaptive gating network to learn optimal fusion weights across different evolutionary stages. Extensive experiments demonstrate that EDS-Net with biological interpretability achieves state-of-the-art performance on both few-shot and zero-shot settings.

Xiaoyue Wan, Xu Zhao

In the context of global population aging, the prevalence of neurodegenerative diseases is rapidly increasing. Vision-based impaired gait analysis emerges as a promising alternative for automatic and non-invasive diagnosis. While prior efforts have advanced either accuracy or interpretability of gait analysis, few have effectively addressed both aspects in a unified framework. To bridge this gap, we propose DPPD, a Diffusion-based Personalized Pathology Disentanglement model that jointly performs quantitative gait scoring, dementia subtyping, and qualitative anomaly highlighting. Motivated by the observation that pathological gait features exhibit stronger inter-class separability across different gait severity than raw features, DPPD is proposed based on the subject-specific pathology disentanglement perspective. Specifically, it comprises three key components: (1) a 3DmotionBERT for encoding gait representation from 3D human pose sequences estimated, (2) a latent diffusion-based Gait Denoiser for generating personalized normal gait features, and (3) a Dual Pathology Disentanglement mechanism that captures both static pose and dynamic motion pathological representation from the residual between raw and normal gait features. These disentangled pathologies further enable quantitative classification and qualitative anomaly highlighting. Experiments on the PDGait and 3DGait datasets demonstrate that DPPD outperforms state-of-the-art methods in classification accuracy while providing reliable and interpretable visualizations of gait anomalies.

Jiaqi Tang, Jianmin Chen, Wei Wei, Xiaogang Xu, Runtao Liu, Xiangyu Wu, Qipeng Xie, Jiafei Wu, Lei Zhang, Qifeng Chen

Multimodal Large Language Models struggle to maintain reliable performance under extreme real-world visual degradations, which impede their practical robustness. Existing robust MLLMs predominantly rely on implicit training/adaptation that focuses solely on visual encoder generalization, suffering from limited interpretability and isolated optimization. To overcome these limitations, we propose Robust-R1, a novel framework that explicitly models visual degradations through structured reasoning chains. Our approach integrates: (i) supervised fine-tuning for degradation-aware reasoning foundations, (ii) reward-driven alignment for accurately perceiving degradation parameters, and (iii) dynamic reasoning depth scaling adapted to degradation intensity. To facilitate this approach, we introduce a specialized 11K dataset featuring realistic degradations synthesized across four critical real-world visual processing stages, each annotated with structured chains connecting degradation parameters, perceptual influence, pristine semantic reasoning chain, and conclusion. Comprehensive evaluations demonstrate state-of-theart robustness: Robust-R1 outperforms all general and robust baselines on the real-world degradation benchmark R-Bench, while maintaining superior anti-degradation performance under multi-intensity adversarial degradations on MMMB, MMStar, and RealWorldQA.

Xiangshuai Song, Jun-Jie Huang, Tianrui Liu, Ke Liang, Chang Tang

Despite the success of convolution- and attention-based models in vision tasks, their rigid receptive fields and complex architectures limit their ability to model irregular spatial patterns and hinder interpretability, thereby posing challenges for tasks requiring high model transparency. Clustering paradigms offer promising interpretability and flexible semantic modeling, but suffer from limited accuracy, low efficiency, and gradient vanishing during training. To address these issues, we propose the CLUster attEntion Network (CLUENet), a transparent deep architecture for visual semantic understanding. Specifically, we introduce three key innovations, including (i) a Global and Soft Feature Aggregation with a Temperature-Scaled Cosine Attention for capturing long-range dependencies and a Gated Fusion Mechanism for enhanced local modeling, (ii) Hard and Shared Feature Dispatching, and (iii) an Improved Cluster Pooling Block. These enhancements significantly improve both classification performance and visual interpretability. Experiments on CIFAR-100 and Mini-ImageNet demonstrate that CLUENet outperforms existing clustering methods and mainstream visual models, offering a compelling balance of accuracy, efficiency, and transparency.

Weimin Shi, Xiang Li, Kaige Li, Junhao Fang, Qiang Zhou, Qichuan Geng, Zhong Zhou

Image geo-localization aims to determine the geographic location of a query image. While Multimodal Large Language Models (MLLMs) show potential for this task due to their rich world knowledge and explainable abilities, they often struggle with confirmation bias, i.e., committing to early, potentially incorrect guesses driven by visual clues with varied geographic likelihoods. In this paper, we propose GeoBayes, a novel training-free framework that formulates geolocalization as a Maximum a Posteriori (MAP) estimation task over multiple geographic hypotheses and performs probabilistic thought via sequential Bayesian reasoning. GeoBayes treats each visual object and its associated geographic clues as probabilistic evidence, integrating them iteratively through a Hypothesize–Verify–Update loop. At each step, it evaluates how new evidence supports existing hypotheses and updates their posterior probabilities, gradually converging on the most probable location. This allows GeoBayes to explicitly quantify and fuse the varied geographic probabilities implied by various visual elements, reducing the risk of overcommitting to misleading clues. Furthermore, considering the natural hierarchy of geographic labels (e.g., country, city), GeoBayes introduces a state memory mechanism that stores hypotheses, inference context, and evidence scores across levels. This design enables the framework to propagate prior knowledge across levels of the geographic hierarchy and incorporate geographic structural constraints into the Bayesian update process, achieving a coarse-to-fine geo-localization. Experiments on IM2GPS3k and YFCC4K show that GeoBayes improves MLLM-based geo-localization accuracy without extra training. This demonstrates the effectiveness of probabilistic reasoning for robust and interpretable geo-localization.

Sahil Shah, S P Sharan, Harsh Goel, Minkyu Choi, Mustafa Munir, Manvik Pasula, Radu Marculescu, Sandeep Chinchali

While vision-language models (VLMs) excel at tasks involving single images or short videos, they still struggle with Long Video Question Answering (LVQA) due to its demand for complex multi-step temporal reasoning. Vanilla approaches, which simply sample frames uniformly and feed them to a VLM along with the question, incur significant token overhead. This forces aggressive downsampling of long videos, causing models to miss fine-grained visual structure, subtle event transitions, and key temporal cues. Recent works attempt to overcome these limitations through heuristic approaches; however, they lack explicit mechanisms for encoding temporal relationships and fail to provide any formal guarantees that the sampled context actually encodes the compositional or causal logic required by the question. To address these foundational gaps, we introduce NeuS-QA, a training-free, plug-and-play neuro-symbolic pipeline for LVQA. NeuS-QA first translates a natural language question into a logic specification that models the temporal relationship between frame-level events. Next, we construct a video automaton to model the video's frame-by-frame event progression, and finally employ model checking to compare the automaton against the specification to identify all video segments that satisfy the question's logical requirements. Only these logic-verified segments are submitted to the VLM, thus improving interpretability, reducing hallucinations, and enabling compositional reasoning without modifying or fine-tuning the model. Experiments on the LongVideoBench and CinePile benchmarks show that NeuS-QA significantly improves performance by over 10%, particularly on questions involving event ordering, causality, and multi-step reasoning.

Fu Rong, Meng Lan, Qian Zhang, Lefei Zhang

Referring Remote Sensing Image Segmentation (RRSIS) aims to segment target objects in remote sensing (RS) images based on textual descriptions. Although Segment Anything Model 2 (SAM2) has shown remarkable performance in various segmentation tasks, its application to RRSIS presents several challenges, including understanding the text-described RS scenes and generating effective prompts from text. To address these issues, we propose RS2-SAM2, a novel framework that adapts SAM2 to RRSIS by aligning the adapted RS features and textual features while providing pseudo-mask-based dense prompts. Specifically, we employ a union encoder to jointly encode the visual and textual inputs, generating aligned visual and text embeddings as well as multimodal class tokens. A bidirectional hierarchical fusion module is introduced to adapt SAM2 to RS scenes and align adapted visual features with the visually enhanced text embeddings, improving the model's interpretation of text-described RS scenes. To provide precise target cues for SAM2, we design a mask prompt generator, which takes the visual embeddings and class tokens as input and produces a pseudo-mask as the dense prompt of SAM2. Experimental results on several RRSIS benchmarks demonstrate that RS2-SAM2 achieves state-of-the-art performance.

Siddharth Roheda, Rohit Chowdhury, Aniruddha Bala, Rohan Jaiswal

We propose a novel Auto-Regressive (AR) image generation approach that models images as hierarchical compositions of interpretable visual layers. While AR models have achieved transformative success in language modeling, replicating this success in vision remains challenging due to inherent spatial dependencies in images. Addressing the unique challenges of vision tasks, our method (CART) adds image details iteratively via semantically meaningful decompositions. We demonstrate the flexibility and generality of CART by applying it across three distinct decomposition strategies: (i) Base-Detail Decomposition (Mumford-Shah smoothness), (ii) Intrinsic Decomposition (albedo/shading), and (iii) Specularity Decomposition (diffuse/specular). This “next-detail" strategy outperforms traditional “next-token" and “next-scale" approaches, improving controllability, semantic interpretability, and resolution scalability. Experiments show CART generates visually compelling results while enabling structured image manipulation, opening new directions for controllable generative modeling via physically or perceptually motivated image factorization.

Ji Rao, Xinyu Liu, Yong Yi, Ying Xiao, Ye Luo

Microvascular invasion (MVI) is a critical prognostic factor that significantly impacts postoperative outcomes in hepatocellular carcinoma (HCC). As the current gold standard for the diagnosis of MVI is based on the postoperative histopathological examination of whole slide images, accurate preoperative prediction of MVI status using magnetic resonance imaging (MRI) presents both a substantial clinical imperative and a significant challenge. In order to discover reliable MRI-based imaging biomarkers to support clinical decision making and enhance the interpretability of deep learning-based diagnostic models, we propose a novel interpretable MVI prediction framework in which the shared latent visual attributes are first learned and then used for potential imaging biomarker extraction and MVI diagnosis, respectively. To ensure that the visual attributes of these biomarkers are generalizable across diverse patients, the similarity constraints at the intra-patient level and the inter-patient level are enforced within the learned feature space, enabling intuitive biomarker discovery directly from the original image space. To guarantee semantic alignment between biomarkers and the characteristics of individual patients, we introduce a novel classification mechanism that directly links the alignment between each biomarker and patient-specific characteristics with the prediction, thereby ensuring a precise prediction of MVI. Furthermore, the interpretability of the model is enhanced by integrating a mask-based visual explanation method that highlights regions in patient images that correspond to the identified biomarkers. Extensive experiments on two MVI prediction datasets: HCC-WCH and HCC-ZSH unequivocally demonstrate our method's superior performance in both classification accuracy and interpretability.