论文检索

Wenbin An, Jiahao Nie, Feng Tian, Mingxiang Cai, Yaqiang Wu, Xiaoqin Zhang, Shijian Lu

Multimodal Retrieval-Augmented Generation (MRAG) has recently been explored to empower Large Vision Language Models (LVLMs) with more comprehensive and up-to-date contextual knowledge, aiming to compensate for their limited and coarse-grained parametric knowledge in knowledge-intensive tasks. However, the retrieved contextual knowledge is usually not aligned with LVLMs’ internal parametric knowledge, leading to knowledge conflicts and further unreliable responses. To tackle this issue, we design KCM, a training-free and plug-and-play framework that can effectively mitigate knowledge conflicts while incorporating MRAG for more accurate LVLM responses. KCM enhances contextual knowledge utilization by modifying the LVLM architecture from three key perspectives. First, KCM adaptively adjusts attention distributions among multiple attention heads, encouraging LVLMs to focus on contextual knowledge with reduced distraction. Second, KCM identifies and prunes knowledge-centric LVLM neurons that encode coarse-grained parametric knowledge, thereby suppressing interferences and enabling more effective integration of contextual knowledge. Third, KCM amplifies the information flow from the input context by injecting supplementary context logits, reinforcing its contribution to the final output. Extensive experiments over multiple LVLMs and benchmarks show that KCM outperforms the state-of-the-art consistently by large margins, incurring neither extra training nor external tools.

Hanrui Chen, Liqi Yan, Qifan Wang, Jianhui Zhang, Fangli Guan, Pan Li

Zero-shot object navigation tasks agents with locating target objects in unseen environments—a core capability of embodied intelligence. While recent vision-language navigation methods leverage Large Language Models (LLMs) for multimodal reasoning, they suffer from two key limitations: (1) semantic misalignment between language-grounded maps and real-world layouts, and (2) inefficiency due to LLMs’ lack of specialization for navigation-specific tasks. To address these challenges, we propose Chain-of-Search (CoS), a novel parameter-efficient framework that enables human-like decision-making via iterative semantic reasoning. First, CoS replaces traditional global maps with an optimal-benefit multi-map construction that continuously balances expected gain and cost throughout the navigation process. Second, we introduce a Parameter-Efficient Intent Aligner (PEIA), trained via a prompt-guided paradigm to align directional decisions with navigation intent. PEIA injects semantic cues into benefit-aware maps, enabling more rational and goal-consistent exploration. Finally, a Reflection-Guided Destination Verifier (RDV) confirms whether the target is reached via language-driven reasoning and corrects potential errors through self-reflection. CoS achieves state-of-the-art performance on HM3D (+2.8% SR) and MP3D (+1.2% SR) without relying on LLMs, demonstrating the effectiveness of lightweight, reasoning-centered navigation.

Jirong Zha, Yuxuan Fan, Tianyu Zhang, Geng Chen, Yingfeng Chen, Chen Gao, Xinlei Chen

Multimodal Large Language Models (MLLMs) have shown promise in single-agent vision tasks, yet benchmarks for evaluating multi-agent collaborative perception remain scarce. This gap is critical, as multi-drone systems provide enhanced coverage, robustness, and collaboration compared to single-sensor setups. Existing multi-image benchmarks mainly target basic perception tasks using high-quality single-agent images, thus failing to evaluate MLLMs in more complex, egocentric collaborative scenarios, especially under real-world degraded perception conditions. To address these challenges, we introduce AirCopBench, the first comprehensive benchmark designed to evaluate MLLMs in embodied aerial collaborative perception under challenging perceptual conditions. AirCopBench includes 14.6k+ questions derived from both simulator and real-world data, spanning four key task dimensions: Scene Understanding, Object Understanding, Perception Assessment, and Collaborative Decision, across 14 task types. We construct the benchmark using data from challenging degraded-perception scenarios with annotated collaborative events, generating large-scale questions through model-, rule-, and human-based methods under rigorous quality control. Evaluations on 40 MLLMs show significant performance gaps in collaborative perception tasks, with the best model trailing humans by 24.38% on average and exhibiting inconsistent results across tasks. Fine-tuning experiments further confirm the feasibility of sim-to-real transfer in aerial collaborative perception.

Yongjun Xiao, Dian Meng, Xinlei Huang, Yanran Liu, Shiwei Ruan, Ziyue Qiao, Xubin Zheng

Effectively modeling multimodal spatial omics data is critical for understanding tissue complexity and underlying biological mechanisms. While spatial transcriptomics, proteomics, and epigenomics capture molecular features, they lack pathological morphological context. Integrating these omics with histopathological images is thus critical for comprehensive disease tissue analysis. However, substantial heterogeneity across omics, imaging, and spatial modalities poses significant challenges. Naive fusion of semantically distinct sources often leads to ambiguous representations. Additionally, the resolution mismatch between high-resolution histology images and lower-resolution sequencing spots complicates spatial alignment. Biological perturbations during sample preparation further distort modality-specific signals, hindering accurate integration. To address these challenges, we propose Graph-guided Representation of Omics and Vision with Expert Regulation for Adaptive Spatial Multi-omics Fusion (GROVER), a novel framework for adaptive integration of spatial multi-omics data. GROVER leverages a Graph Convolutional Network encoder based on Kolmogorov–Arnold Networks to capture the nonlinear dependencies between each modality and its associated spatial structure, thereby producing expressive, modality-specific embeddings. To align these representations, we introduce a spot-feature-pair contrastive learning strategy that explicitly optimizes the correspondence across modalities at each spot. Furthermore, we design a dynamic expert routing mechanism that adaptively selects informative modalities for each spot while suppressing noisy or low-quality inputs. Experiments on real-world spatial omics datasets demonstrate that GROVER outperforms state-of-the-art baselines, providing a robust and reliable solution for multimodal integration.

Wenshuo Wang, Yaomin Shen, Yingjie Tan, Yihao Chen

Spatiotemporal forecasting often relies on computationally intensive models to capture complex dynamics. Knowledge distillation (KD) has emerged as a key technique for creating lightweight student models, with recent advances like frequency-aware KD successfully preserving spectral properties (i.e., high-frequency details and low-frequency trends). However, these methods are fundamentally constrained by operating on pixel-level signals, leaving them blind to the rich semantic and causal context behind the visual patterns. To overcome this limitation, we introduce S2-KD, a novel framework that unifies Semantic priors with Spectral representations for distillation. Our approach begins by training a privileged, multimodal teacher model. This teacher leverages textual narratives from a Large Multimodal Model (LMM) to reason about the underlying causes of events, while its architecture simultaneously decouples spectral components in its latent space. The core of our framework is a new distillation objective that transfers this unified semantic-spectral knowledge into a lightweight, vision-only student. Consequently, the student learns to make predictions that are not only spectrally accurate but also semantically coherent, without requiring any textual input or architectural overhead at inference. Extensive experiments on benchmarks like WeatherBench and TaxiBJ+ show that S2-KD significantly boosts the performance of simple student models, enabling them to outperform state-of-the-art methods, particularly in long-horizon and complex non-stationary scenarios.

Jiaheng Wang, Yuan Si, Ang Li, Zhenyu Wang, Tianheng Xu, Honglin Hu

Precise detection of driver mental fatigue is critical for reducing traffic accidents and enhancing road safety. Compared with vision-based detection—which is susceptible to illumination and occlusion—multimodal physiological‑signal-based approaches integrate complementary information from diverse biosignals, delivering more faithful and objective fatigue assessments. However, adverse factors such as motion artifacts and environmental noise induce ceaseless deterioration to physiological signals, which markedly degrade the performance of existing multimodal fusion methods. To address this challenge, we propose Multimodal Uncertainty-based Self-driven Evolution, MUSE, reallocating modality contributions in real time via overall uncertainty minimization, thereby enabling efficient collaborative fusion of multi‐source predictions. Theoretically, MUSE guarantees a provably bounded cumulative error, and its generalization error approaches the Bayesian‑optimal fusion as iterations progress. Operating in a closed loop without labels or manual recalibration, MUSE presents superior suitability for real‑world driving scenarios compared to supervised algorithms. On the large‑scale driving fatigue dataset SEED‑VIG, MUSE outperforms existing models in both classification and regression tasks, substantiating its robustness and practicality as a promising driving fatigue detection solution.

Fanxiao Li, Jiaying Wu, Tingchao Fu, Yunyun Dong, Bingbing Song, Wei Zhou

The proliferation of multimodal misinformation poses growing threats to public discourse and societal trust. While Large Vision-Language Models (LVLMs) have enabled recent progress in multimodal misinformation detection (MMD), the rise of generative AI (GenAI) tools introduces a new challenge: GenAI-driven news diversity, characterized by highly varied and complex content. We show that this diversity induces multi-level drift, comprising (1) model-level misperception drift, where stylistic variations disrupt a model’s internal reasoning, and (2) evidence-level drift, where expression diversity degrades the quality or relevance of retrieved external evidence. These drifts significantly degrade the robustness of current LVLM-based MMD systems. To systematically study this problem, we introduce DriftBench, a large-scale benchmark comprising 16,000 news instances across six categories of diversification. We design three evaluation tasks: (1) robustness of truth verification under multi-level drift; (2) susceptibility to adversarial evidence contamination generated by GenAI; and (3) analysis of reasoning consistency across diverse inputs. Experiments with six state-of-the-art LVLM-based detectors show substantial performance drops (average F1 ↓ 14.8%) and increasingly unstable reasoning traces, with even more severe failures under adversarial evidence injection. Our findings uncover fundamental vulnerabilities in existing MMD systems and suggest an urgent need for more resilient approaches in the GenAI era.

Zongsheng Cao, Anran Liu, Yangfan He, Jing Li, Bo Zhang, Zigan Wang

Retrieval-augmented generation (RAG) has greatly improved Large Language Models (LLMs) by adding external knowledge. However, current RAG-based methods face difficulties with long-context video understanding due to two main challenges. First, Current RAG-based methods for long-context video understanding struggle to effectively integrate multimodal and long-range temporal information, resulting in fragmented and context-insensitive knowledge representations. Furthermore, their retrieval mechanisms often rely on static textual matching, failing to dynamically align user queries with the most relevant video segments and leading to suboptimal downstream performance. To overcome these issues, we introduce ViG-RAG, a new framework to enhance long-context video understanding through structured textual knowledge grounding and multi-modal retrieval. Specifically, we segment video transcripts into structured units, extract key entities, form temporal connections, and assign confidence for evidence, enabling coherent long-range reasoning. In this way, it utilizes a knowledge-aware grounding mechanism and a context-aware retrieval process that dynamically builds a probabilistic temporal knowledge graph to organize multi-video content. To improve retrieval accuracy, we propose a hybrid retrieval strategy for semantic and temporal features, with an adaptive distribution modeling the relevance. In this way, it achieves the optimal retrieval distribution for each query, enhancing generation efficiency by reducing unnecessary computations. On top of this, ViG-RAG uses a vision-language model to integrate semantic anchors, expanded contextual fields, and selected video frames, generating an accurate response. We evaluate ViG-RAG on several benchmarks, demonstrating that it significantly surpasses current RAG-based methods.

Jiahao Yang, Zihan Wang, Xiangyang Li, Xing Zhu, Yujun Shen, Yinghao Xu, Shuqiang Jiang

Despite significant progress in Vision-Language Navigation (VLN), existing approaches still rely on dense RGB videos that produce excessive patch tokens and lack explicit spatial structure, resulting in substantial computational overhead and limited spatial reasoning. To address these issues, we introduce the Geometry-Aware BEV (GA-BEV) --a compact, 3D-grounded feature representation that integrates both explicit and implicit geometric cues into multimodal large language model (MLLM)-based navigation systems. We construct BEV spatial maps from RGB-D inputs by projecting visual features into 3D space and aggregating them into an agent-centric layout that preserves geometric consistency while reducing token redundancy. To further enrich geometric understanding, we incorporate features from a pretrained 3D foundation model into the BEV space, injecting structural priors learned from large-scale 3D reconstruction tasks. Together, these complementary cues--explicit depth-based projection and implicit learned priors--yield compact yet spatially expressive representations that substantially improve navigation efficiency and performance. Experiments show that our method achieves state-of-the-art results using only navigation data, without DAgger augmentation or mixed VQA training, demonstrating the robustness and data efficiency of the proposed GA-VLN framework.

Soroush Seifi, Simon Gardier, Vaggelis Dorovatas, Daniel Olmeda Reino, Rahaf Aljundi

AI assistants that support humans in daily life are becoming increasingly feasible, driven by the rapid advancements in multimodal language models. A key challenge lies in overcoming the generic nature of these models to deliver personalized experiences. Existing approaches to personalizing large vision language models often rely on additional training stages, which limit generality and scalability, or on engineered pipelines with external pre-trained modules, which hinder deployment efficiency. In this work, we propose an efficient personalization method that leverages the model's inherent ability to capture personalized concepts. Specifically, we extract visual tokens that predominantly represent the target concept by utilizing the model's internal attention mechanisms. These tokens serve as a memory of that specific concept, enabling the model to recall and describe it when it appears in test images. We conduct a comprehensive and unified evaluation of our approach and SOTA methods across various personalization settings including single-concept, multi-concept, and video personalization, demonstrating strong performance gains with minimal personalization overhead.

Haosong Peng, Hao Li, Yalun Dai, Yushi Lan, Yihang Luo, Tianyu Qi, Zhengshen Zhang, Yufeng Zhan, Junfei Zhang, Wenchao Xu 等

General 3D foundation models have started to lead the trend of unifying diverse vision tasks, yet most assume RGB-only inputs and ignore readily available geometric cues (e.g., camera intrinsics, poses, and depth maps). To address this issue, we introduce OmniVGGT, a novel framework that can effectively benefit from an arbitrary number of auxiliary geometric modalities during both training and inference. In our framework, a GeoAdapter is proposed to encode depth and camera intrinsics/extrinsics into a spatial foundation model. It employs zero-initialized convolutions to progressively inject geometric information without disrupting the foundation model's representation space. This design ensures stable optimization with negligible overhead, maintaining inference speed comparable to VGGT even with multiple additional inputs. Additionally, a stochastic multimodal fusion regimen is proposed, which randomly samples modality subsets per instance during training. This enables an arbitrary number of modality inputs during testing and promotes learning robust spatial representations instead of overfitting to auxiliary cues. Extensive experiments on monocular/multi-view depth estimation, multi-view stereo, and camera pose estimation demonstrate that OmniVGGT outperforms prior methods with auxiliary inputs and achieves state-of-the-art results even with RGB-only input. To further highlight its practical utility, we integrated OmniVGGT into vision-language-action (VLA) models. The enhanced VLA model by OmniVGGT not only outperforms the vanilla point-cloud-based baseline on mainstream benchmarks, but also effectively leverages accessible auxiliary inputs to achieve consistent gains on robotic tasks. Project Page: https://livioni.github.io/OmniVGGT-official/

Chenyang Gu, Jiaming Liu, Hao Chen, Runzhong Huang, Qingpo Wuwu, Xiaoqi Li, Zhuoyang Liu, Ying Li, Renrui Zhang, Peng Jia 等

Vision-Language-Action (VLA) models have recently emerged, demonstrating strong generalization in robotic scene understanding and manipulation. However, when confronted with long-horizon tasks that require defined goal states, such as LEGO assembly or object rearrangement, existing VLA models still face challenges in coordinating long-horizon planning with precise manipulation.Therefore, we aim to endow a VLA model with the capability to infer the "how" process from the "what" outcomes, transforming goal states into executable procedures. In this paper, we introduce ManualVLA, a unified VLA framework built upon a Mixture-of-Transformers (MoT) architecture, enabling coherent collaboration between multimodal manual generation and action execution. Unlike prior VLA models that directly map sensory inputs to actions, we first equip ManualVLA with a planning expert that generates intermediate manuals consisting of images, visual prompts, and textual instructions. Building upon these multimodal manuals, we design a Manual Chain-of-Thought (ManualCoT) reasoning process that feeds them into the action expert, where each manual step provides explicit control conditions, while its latent representation offers implicit guidance for accurate manipulation. To alleviate the burden of data collection, we develop a high-fidelity digital-twin toolkit based on 3D Gaussian Splatting, which automatically generates manual data for planning expert training. ManualVLA demonstrates strong real-world performance, achieving an average success rate 32% higher than the previous hierarchical SOTA baseline on LEGO assembly and object rearrangement tasks.

Guolei Huang, Qinzhi Peng, Gan Xu, Yao Huang, Yuxuan Lu, Yongjun Shen

As Vision-Language Models (VLMs) move into interactive, multi-turn use, safety concerns intensify for multimodal multi-turn dialogue, which is characterized by concealment of malicious intent, contextual risk accumulation, and cross-modal joint risk. These characteristics limit the effectiveness of content moderation approaches designed for single-turn or single-modality settings. To address these limitations, we first construct the Multimodal Multi-turn Dialogue Safety (MMDS) dataset, comprising 4,484 annotated dialogues and a comprehensive risk taxonomy with 8 primary and 60 subdimensions. As part of MMDS construction, we introduce Multimodal Multi-turn Red Teaming (MMRT), an automated framework for generating unsafe multimodal multi-turn dialogues. We further propose LLaVAShield, which audits the safety of both user inputs and assistant responses under specified policy dimensions in multimodal multi-turn dialogues. Extensive experiments show that LLaVAShield significantly outperforms state-of-the-art VLMs and existing content moderation tools while demonstrating strong generalization and flexible policy adaptation. Additionally, we analyze vulnerabilities of mainstream VLMs to harmful inputs and evaluate the contribution of key components, advancing understanding of safety mechanisms in multimodal multi-turn dialogues.

Haoyu He, Yue Zhuo, Yu Zheng, Qi R. Wang

Vision-language models (VLMs) achieve strong multimodal performance, yet how computation is organized across populations of neurons remains poorly understood. In this work, we study VLMs through the lens of neural topology, representing each layer as a within-layer correlation graph derived from neuron-neuron co-activations. This view allows us to ask whether population-level structure is behaviorally meaningful, how it changes across modalities and depth, and whether it identifies causally influential internal components under intervention. We show that correlation topology carries recoverable behavioral signal; moreover, cross-modal structure progressively consolidates with depth around a compact set of recurrent hub neurons, whose targeted perturbation materially alters model output. Neural topology thus emerges as a meaningful intermediate scale for VLM interpretability: richer than local attribution, more tractable than full circuit recovery, and empirically tied to multimodal behavior. Code is publicly available at https://github.com/he-h/vlm-graph-probing.

Jiaxu Wan, Xu Wang, Mengwei Xie, Hang Zhang, Mu Xu, Yang Han, Ding Yuan, Hong Zhang, Yifan Yang

Video-based spatial reasoning -- such as estimating distances, judging directions, or understanding layouts from multiple views -- requires selecting informative frames and, when needed, actively seeking additional viewpoints during inference. Existing multimodal large language models (MLLMs) consume a fixed set of uniformly sampled frames and cannot request new views once reasoning begins, often missing the geometric cues necessary for reliable spatial judgments. We present EagleVision, a dual-stage framework that combines geometry-aware frame selection with active, Bird's-Eye-View (BEV)-grounded reasoning. In the first stage (macro perception), a semantics-perspective-fusion determinantal point process (SPF-DPP) selects a compact set of keyframes that jointly maximize semantic relevance and viewpoint diversity under a fixed token budget. In the second stage (micro verification), the model performs iterative spatial Chain-of-Thought: at each step it can either reason in text or predict a pose on the BEV plane to retrieve the nearest real frame, forming a closed-loop hypothesize-look-verify cycle. The querying policy is trained purely via reinforcement learning with a spatial grounding reward, requiring no human-annotated reasoning traces. On VSI-Bench and SQA3D, EagleVision achieves state-of-the-art performance among open-source vision-language models.

Jovana Kondic, Pengyuan Li, Dhiraj Joshi, Isaac Sanchez, Ben Wiesel, Shafiq Abedin, Amit Alfassy, Eli Schwartz, Daniel Caraballo, Yagmur Gizem Cinar 等

Understanding charts requires models to jointly reason over geometric visual patterns, structured numerical data, and natural language -- a capability where current vision-language models (VLMs) remain limited. We introduce ChartNet, a high-quality, million-scale multimodal dataset designed to advance chart interpretation and reasoning. ChartNet leverages a novel code-guided synthesis pipeline to generate 1.5 million diverse chart samples spanning 24 chart types and 6 plotting libraries. Each sample consists of five aligned components: plotting code, rendered chart image, data table, natural language summary, and question-answering with reasoning, providing fine-grained cross-modal alignment. To capture the full spectrum of chart comprehension, ChartNet additionally includes specialized subsets encompassing human annotated data, real-world data, safety, and grounding. Moreover, a rigorous quality-filtering pipeline ensures visual fidelity, semantic accuracy, and diversity across chart representations. Fine-tuning on ChartNet consistently improves results across benchmarks, demonstrating its utility as large-scale supervision for multimodal models. As the largest open-source dataset of its kind, ChartNet aims to support the development of foundation models with robust and generalizable capabilities for data visualization understanding. The dataset is publicly available at https://huggingface.co/datasets/ibm-granite/ChartNet

Abdul Rehman, Iqra Rasool, Ayisha Imran, Mohsen Ali, Waqas Sultani

Digital hematopathology requires cell-level analysis across diverse disease categories, including malignant disorders (e.g., leukemia), infectious conditions (e.g., malaria), and non-malignant red blood cell disorders (e.g., sickle cell disease). Whether single-task, vision-language, WSI- optimized, or single-cell hematology models, these approaches share a key limitation: they cannot provide unified, multi-task, multi-modal reasoning across the complexities of digital hematopathology. To overcome these limitations, we propose Uni-Hema, a multi-task, unified model for digital hematopathology integrating detection, classification, segmentation, morphology prediction, and reasoning across multiple diseases. Uni-Hema leverages 46 publicly available datasets, encompassing over 700K images and 21K question-answer pairs, and is built upon Hema-Former, a multimodal module that bridges visual and linguistic representations at the hierarchy level for the different tasks (detection, classification, segmentation, morphology, mask language modeling, and visual question answering) at different granularities. Extensive experiments demonstrate that Uni-Hema achieves comparable or superior performance compared to training on a single task and single-dataset models, across diverse hematological tasks, while providing interpretable, morphologically relevant insights at the single-cell level. Our framework establishes a new standard for multi-task and multi-modal digital hematopathology. The code is available at https://github.com/intelligentMachines-ITU/Uni-Hema

Mujtaba Hussain Mirza, Antonio D'Orazio, Odelia Melamed, Iacopo Masi

Despite the rapid progress in multimodal models and Large Visual-Language Models (LVLM), they remain highly susceptible to adversarial perturbations, raising serious concerns about their reliability in real-world use. While adversarial training has become the leading paradigm for building models that are robust to adversarial attacks, Test-Time Transformations (TTT) have emerged as a promising strategy to boost robustness at inference.In light of this, we propose Energy-Guided Test-Time Transformation (ET3), a lightweight, training-free defense that enhances the robustness by minimizing the energy of the input samples.Our method is grounded in a theory that proves our transformation succeeds in classification under reasonable assumptions. We present extensive experiments demonstrating that ET3 provides a strong defense for classifiers, zero-shot classification with CLIP, and also for boosting the robustness of LVLMs in tasks such as Image Captioning and Visual Question Answering. Code is available at github.com/OmnAI-Lab/Energy-Guided-Test-Time-Defense .

Zhiheng Wu, Tong Wang, Shuning Wang, Naiming Liu, Yumeng Zhang

Recent advances in Vision-Language Models (VLMs) have benefited from Reinforcement Learning (RL) for enhanced reasoning. However, existing methods still face critical limitations, including the lack of low-level visual information and effective visual feedback. To address these problems, this paper proposes a unified multimodal interleaved reasoning framework ForeSight, which enables VLMs to See Further with low-level visual cues and Think Deeper with effective visual feedback. First, it introduces a set of low-level visual tools to integrate essential visual information into the reasoning chain, mitigating the neglect of fine-grained visual features. Second, a mask-based visual feedback mechanism is elaborated to incorporate visual reflection into the thinking process, enabling the model to dynamically re-examine and update its answers. Driven by RL, ForeSight learns to autonomously decide on tool invocation and answer verification, with the final answer accuracy as the reward signal. To evaluate the performance of the proposed framework, we construct a new dataset, Character and Grounding SalBench (CG-SalBench), based on the SalBench dataset. Experimental results demonstrate that the ForeSight-7B model significantly outperforms other models with the same parameter scale, and even surpasses the current SOTA closed-source models on certain metrics.

Jingxuan Wei, Caijun Jia, Xi Bai, Xinglong Xu, Siyuan Li, Linzhuang Sun, Bihui Yu, Conghui He, Lijun Wu, Cheng Tan

Unified Multimodal Models (UMMs) are redefining the landscape of artificial intelligence by coupling perception and generation across language, vision, and structured reasoning. Yet, despite their growing sophistication, a critical gap persists in evaluation: existing benchmarks largely measure discriminative understanding or unconstrained generation in isolation, overlooking the integrated generative reasoning required for genuine multimodal intelligence. To address this, we introduce GGBench, the benchmark explicitly designed to evaluate geometric generative reasoning--the ability of a model to understand, reason about, and construct a solution within a unified framework. Each instance in GGBench contains precisely aligned natural-language instructions, executable GeoGebra code, and rendered diagrams, enabling deterministic and interpretable verification of a model's reasoning and constructive fidelity. The benchmark comprises 1,411 rigorously curated problems covering eight categories and multiple difficulty levels, resulting in over 7,000 aligned visualizations. We propose a comprehensive tri-modal evaluation protocol that jointly assesses textual planning quality, code executability, and geometric accuracy of generated diagrams through both automated and human-in-the-loop judging. Extensive experiments on both state-of-the-art UMMs and general Large Language Models (LLMs) reveal a large performance gap between end-to-end generation and reasoning-grounded construction. GGBench establishes a new standard for testing multimodal systems that must not only understand but also build, marking a crucial step toward grounded, verifiable generative intelligence.