大华星汉大模型系列让智能体从“看得清”到“自主认知”，展现CV×AI Agent在行业场景中的落地新范式。
让智能体“看见”世界，不是给它装上摄像头那么简单。当CV遇上AI Agent，乘数效应正在发生。AI Agent不等于Chatbot，真正的价值在于静默看护与自主决策。 AI进入产业腹地仍面临两大鸿沟：视觉认知与行业的理解。为此，大华股份构建了星汉大模型系列——V系列视觉大模型让系统“看得懂”物理世…
多模态大模型多人图像输入的安全漏洞：组合攻击框架DMN揭示新的越狱风险。
arXiv:2605.18915v1 Announce Type: cross Abstract: Multimodal Large Language Models (MLLMs) are vulnerable to jailbreak attacks, which can elicit harmf…
探索指令微调多模态大模型在自然刺激下的脑区对齐模式，交叉验证AI与神经科学
arXiv:2506.08277v3 Announce Type: replace-cross Abstract: Recent voxel-wise multimodal brain encoding studies have shown that multimodal large languag…
多模态大模型新突破：学会判断何时开口，对话交互更自然
arXiv:2505.14654v2 Announce Type: replace-cross Abstract: Chatbots via large language models (LLMs) generate fluent responses but often struggle with …
新基准Artifact-Bench，专门评估多模态大模型识别AI视频伪影的能力，揭示模型真实短板。
arXiv:2605.18984v1 Announce Type: new Abstract: Recent video generative models have greatly improved the realism of AI-generated videos, yet their out…
StrLoRA提出流式持续视觉指令微调新方法，有效缓解多模态大模型在序列任务中的灾难性遗忘。
arXiv:2605.16353v1 Announce Type: new Abstract: Continual Visual Instruction Tuning (CVIT) enables Multimodal Large Language Models to incrementally a…
从视频中自动合成海量GUI交互轨迹，破解GUI Agent预训练数据稀缺难题，让智能体更好理解真实应用。
arXiv:2605.14747v1 Announce Type: cross Abstract: Recent advances in multimodal large language models have driven growing interest in graphical user i…
多模态大模型训练新范式：阶段感知稀疏性动态消除冗余，大幅提升效率而保持性能。
arXiv:2509.18150v2 Announce Type: replace Abstract: Multimodal Large Language Models (MLLMs) have demonstrated outstanding performance across a variet…
最新诊断论文深入评估多模态大模型在全景图像中的视角条件空间推理能力，超越传统定位局限
arXiv:2605.12413v3 Announce Type: replace Abstract: Multimodal Large Language Models (MLLMs) show strong visual perception, yet remain limited in reas…
研究发现视频MLLMs的音频理解实际依赖视觉线索，揭示模型幻觉问题，挑战多模态真实性。
arXiv:2605.16403v1 Announce Type: new Abstract: Despite rapid progress in video-capable MLLMs, we find that their apparent audio understanding in vide…
研究发现多模态大模型在临床序数评分中易出现中心趋势偏差，影响评估准确性。
arXiv:2605.16386v1 Announce Type: new Abstract: Multimodal large language models (LLMs) are increasingly explored as automated evaluators in clinical …
提出频域残差压缩方法，大幅减少视频MLLM的token数量，高效且不损失性能。
arXiv:2605.16366v1 Announce Type: new Abstract: Video MLLMs face a persistent tension between spatial fidelity and temporal coverage: preserving fine-…
多模态大模型新突破，通过自蒸馏策略让AI学会捕捉视觉细节，显著提升细粒度理解能力。
arXiv:2605.18740v1 Announce Type: cross Abstract: Multimodal Large Language Models (MLLMs) still struggle with fine-grained visual understanding, wher…
用方程锚定工具使用，让多模态大模型实现相机鲁棒的3D定位，突破视觉视角限制。
arXiv:2605.19528v1 Announce Type: new Abstract: 3D localization in Multimodal Large Language Models (MLLMs), including 3D object detection and 3D visu…
用数据集+模型+基准全方位提升多模态大模型跨视图空间智能，突破单视角局限。
arXiv:2605.18621v1 Announce Type: new Abstract: Spatial intelligence requires multimodal large language models (MLLMs) to move beyond single-view perc…
多模态大模型在空间智能上的突破，赋予AI更强的视觉感知与推理能力。
arXiv:2505.23747v2 Announce Type: replace-cross Abstract: Recent advancements in Multimodal Large Language Models (MLLMs) have significantly enhanced …
揭示视觉指令调优中样本信息性的关键作用，COMPACT方法用少量高信息样本高效微调多模态大模型
arXiv:2504.21850v3 Announce Type: replace Abstract: Visual instruction tuning (VIT) datasets have grown rapidly in scale, yet the informativeness of i…
提出检索与排序增强多模态大模型，精准攻克CLIP细粒度视觉识别短板
arXiv:2403.13805v2 Announce Type: replace-cross Abstract: CLIP (Contrastive Language-Image Pre-training) uses contrastive learning from noise image-te…