cs.CV（2026-03-31）

📊 共 38 篇论文 | 🔗 7 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (13 🔗2) 支柱三：空间感知与语义 (Perception & Semantics) (10 🔗3) 支柱二：RL算法与架构 (RL & Architecture) (5) 支柱一：机器人控制 (Robot Control) (4) 支柱四：生成式动作 (Generative Motion) (2 🔗1) 支柱六：视频提取与匹配 (Video Extraction) (2) 支柱五：交互与反应 (Interaction & Reaction) (1 🔗1) 支柱七：动作重定向 (Motion Retargeting) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (13 篇)

#题目一句话要点标签🔗⭐
1 Adversarial Prompt Injection Attack on Multimodal Large Language Models 提出针对多模态大语言模型的不可察觉视觉提示注入攻击方法 large language model multimodal instruction following
2 Scaling the Long Video Understanding of Multimodal Large Language Models via Visual Memory Mechanism 提出FlexMem，通过视觉记忆机制增强多模态大语言模型对长视频的理解能力 large language model multimodal
3 Less Is More? Selective Visual Attention to High-Importance Regions for Multimodal Radiology Summarization ViTAS：通过选择性关注重要区域，提升多模态放射学报告摘要生成效果 multimodal
4 Unify-Agent: A Unified Multimodal Agent for World-Grounded Image Synthesis 提出Unify-Agent，用于解决世界知识驱动图像合成中长尾概念生成难题。 multimodal
5 Few-shot Writer Adaptation via Multimodal In-Context Learning 提出基于多模态上下文学习的少样本手写体作者自适应方法 multimodal
6 Multimodal Models Meet Presentation Attack Detection on ID Documents 利用多模态模型进行身份证件的呈现攻击检测研究 multimodal
7 Developing Adaptive Context Compression Techniques for Large Language Models (LLMs) in Long-Running Interactions 提出自适应上下文压缩框架，解决LLM长程交互中的性能退化问题 large language model
8 EC-Bench: Enumeration and Counting Benchmark for Ultra-Long Videos EC-Bench：用于超长视频枚举和计数的基准测试，挑战多模态大语言模型。 large language model multimodal
9 CutClaw: Agentic Hours-Long Video Editing via Music Synchronization CutClaw：提出基于音乐同步的多智能体框架，实现小时级视频的自动剪辑。 multimodal ✅
10 MELT: Improve Composed Image Retrieval via the Modification Frequentation-Rarity Balance Network 提出MELT网络，通过平衡频率和稀有性来改进组合图像检索。 multimodal ✅
11 Omni-NegCLIP: Enhancing CLIP with Front-Layer Contrastive Fine-Tuning for Comprehensive Negation Understanding Omni-NegCLIP通过前层对比微调增强CLIP对否定语句的理解能力 multimodal
12 Diffusion Mental Averages 提出Diffusion Mental Averages (DMA)，在扩散模型中生成概念的清晰逼真“心理平均”图像。 multimodal
13 LatentPilot: Scene-Aware Vision-and-Language Navigation by Dreaming Ahead with Latent Visual Reasoning LatentPilot：利用潜在视觉推理进行前瞻性规划的场景感知视觉-语言导航 VLN

🔬 支柱三：空间感知与语义 (Perception & Semantics) (10 篇)

#题目一句话要点标签🔗⭐
14 LightHarmony3D: Harmonizing Illumination and Shadows for Object Insertion in 3D Gaussian Splatting LightHarmony3D：在3D高斯溅射中实现光照和阴影协调的对象插入 3D gaussian splatting 3DGS gaussian splatting
15 AA-Splat: Anti-Aliased Feed-forward Gaussian Splatting 提出AA-Splat，解决FF-3DGS在新视角合成中因采样率变化导致的渲染伪影问题 3D gaussian splatting 3DGS gaussian splatting
16 Hierarchical Visual Relocalization with Nearest View Synthesis from Feature Gaussian Splatting 提出SplatHLoc，利用特征高斯溅射进行分层视觉重定位，提升鲁棒性。 gaussian splatting splatting feature matching
17 MotionScale: Reconstructing Appearance, Geometry, and Motion of Dynamic Scenes with Scalable 4D Gaussian Splatting MotionScale：通过可扩展的4D高斯溅射重建动态场景的外观、几何和运动 gaussian splatting splatting ✅
18 ConInfer: Context-Aware Inference for Training-Free Open-Vocabulary Remote Sensing Segmentation 提出ConInfer，通过上下文感知推理解决遥感图像的免训练开放词汇分割问题。 open-vocabulary open vocabulary ✅
19 StereoVGGT: A Training-Free Visual Geometry Transformer for Stereo Vision 提出StereoVGGT，一种免训练的立体视觉几何Transformer，显著提升立体匹配性能。 depth estimation monocular depth VGGT
20 Hallucination-aware intermediate representation edit in large vision-language models 提出幻觉感知中间表示编辑框架，有效消除大型视觉语言模型中的幻觉问题。 scene understanding multimodal ✅
21 GRVS: a Generalizable and Recurrent Approach to Monocular Dynamic View Synthesis 提出GRVS：一种通用且循环的单目动态视角合成方法 gaussian splatting splatting
22 Extend3D: Town-Scale 3D Generation Extend3D：提出一种基于单张图像的城市级3D场景生成无训练流程。 monocular depth
23 M2H-MX: Multi-Task Dense Visual Perception for Real-Time Monocular Spatial Understanding M2H-MX：用于实时单目空间理解的多任务密集视觉感知模型 semantic map

🔬 支柱二：RL算法与架构 (RL & Architecture) (5 篇)

#题目一句话要点标签🔗⭐
24 Assessing Multimodal Chronic Wound Embeddings with Expert Triplet Agreement TriDerm：利用专家三重比较评估多模态慢性伤口嵌入，提升罕见皮肤病相似病例检索。 representation learning large language model foundation model
25 SceneTeract: Agentic Functional Affordances and VLM Grounding in 3D Scenes SceneTeract：提出具身智能体功能可供性和VLM在3D场景中的对齐框架 distillation scene understanding affordance
26 Scaling Video Pretraining for Surgical Foundation Models SurgRec：可扩展的手术视频预训练模型，用于构建手术领域的基础模型。 JEPA MAE foundation model
27 Quantization with Unified Adaptive Distillation to enable multi-LoRA based one-for-all Generative Vision Models on edge 提出QUAD框架，实现边缘设备上多LoRA自适应的生成视觉模型量化与部署。 distillation foundation model
28 Square Superpixel Generation and Representation Learning via Granular Ball Computing 提出基于粒计算的方形超像素生成与表征学习方法，提升视觉任务性能。 representation learning

🔬 支柱一：机器人控制 (Robot Control) (4 篇)

#题目一句话要点标签🔗⭐
29 MaskAdapt: Learning Flexible Motion Adaptation via Mask-Invariant Prior for Physics-Based Characters 提出MaskAdapt以解决物理基础角色的灵活运动适应问题 humanoid humanoid control motion adaptation
30 FED-Bench: A Cross-Granular Benchmark for Disentangled Evaluation of Facial Expression Editing 提出FED-Bench，用于解耦评估面部表情编辑的跨粒度基准。 manipulation instruction following
31 Seeing the Evidence, Missing the Answer: Tool-Guided Vision-Language Models on Visual Illusions 提出工具引导的视觉-语言模型框架，解决VLM在视觉错觉识别中的系统性偏差问题。 manipulation
32 A2BFR: Attribute-Aware Blind Face Restoration 提出A$^2$BFR框架，实现属性可控的高保真盲人脸修复。 manipulation

🔬 支柱四：生成式动作 (Generative Motion) (2 篇)

#题目一句话要点标签🔗⭐
33 Not All Frames Are Equal: Complexity-Aware Masked Motion Generation via Motion Spectral Descriptors 提出DynMask，通过运动谱描述器实现复杂度感知的掩码运动生成。 text-to-motion motion synthesis motion generation ✅
34 Emotion Diffusion Classifier with Adaptive Margin Discrepancy Training for Facial Expression Recognition 提出自适应边缘差异训练的情感扩散分类器，提升面部表情识别的鲁棒性。 motion diffusion

🔬 支柱六：视频提取与匹配 (Video Extraction) (2 篇)

#题目一句话要点标签🔗⭐
35 PRISM: A Multi-View Multi-Capability Retail Video Dataset for Embodied Vision-Language Models PRISM：用于具身视觉-语言模型的多视角零售视频数据集 egocentric chain-of-thought
36 Storing Less, Finding More: How Novelty Filtering Improves Cross-Modal Retrieval on Edge Cameras 提出新颖性过滤方法，提升边缘相机跨模态检索性能 egocentric

🔬 支柱五：交互与反应 (Interaction & Reaction) (1 篇)

#题目一句话要点标签🔗⭐
37 Leveraging Synthetic Data for Enhancing Egocentric Hand-Object Interaction Detection 利用合成数据增强以自我为中心的视角下人-物交互检测 HOI egocentric ✅

🔬 支柱七：动作重定向 (Motion Retargeting) (1 篇)

#题目一句话要点标签🔗⭐
38 SkeletonContext: Skeleton-side Context Prompt Learning for Zero-Shot Skeleton-based Action Recognition SkeletonContext：利用骨骼上下文提示学习实现零样本骨骼动作识别 motion representation

⬅️ 返回 cs.CV 首页 · 🏠 返回主页