cs.CV（2025-12-22）

📊 共 26 篇论文 | 🔗 6 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (10 🔗1) 支柱三：空间感知与语义 (Perception & Semantics) (5 🔗3) 支柱二：RL算法与架构 (RL & Architecture) (4 🔗1) 支柱一：机器人控制 (Robot Control) (3) 支柱六：视频提取与匹配 (Video Extraction) (2) 支柱四：生成式动作 (Generative Motion) (1 🔗1) 支柱八：物理动画 (Physics-based Animation) (1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (10 篇)

#题目一句话要点标签🔗⭐
1 Generative Giants, Retrieval Weaklings: Why do Multimodal Large Language Models Fail at Multimodal Retrieval? 揭示多模态大语言模型在多模态检索中表现不佳的原因 large language model multimodal
2 Multimodal LLMs for Historical Dataset Construction from Archival Image Scans: German Patents (1877-1918) 利用多模态LLM从档案图像扫描件构建德国专利历史数据集 large language model multimodal
3 ReasonCD: A Multimodal Reasoning Large Model for Implicit Change-of-Interest Semantic Mining ReasonCD：提出多模态推理大模型，用于遥感图像隐式兴趣变更语义挖掘 large language model multimodal
4 Beyond CLIP: Knowledge-Enhanced Multimodal Transformers for Cross-Modal Alignment in Diabetic Retinopathy Diagnosis 提出知识增强多模态Transformer，用于糖尿病视网膜病变诊断中的跨模态对齐 multimodal
5 dMLLM-TTS: Self-Verified and Efficient Test-Time Scaling for Diffusion Multi-Modal Large Language Models 提出dMLLM-TTS，通过自验证和高效测试时缩放提升扩散多模态大语言模型的生成质量和效率。 large language model ✅
6 From Indoor to Open World: Revealing the Spatial Reasoning Gap in MLLMs 揭示多模态大语言模型在开放世界空间推理方面的差距，并提出相应评测基准。 large language model multimodal
7 MapTrace: Scalable Data Generation for Route Tracing on Maps MapTrace：提出可扩展数据生成流程，提升MLLM地图路径追踪能力 large language model multimodal
8 D2Pruner: Debiased Importance and Structural Diversity for MLLM Token Pruning D2Pruner：通过解偏重要性和结构多样性实现MLLM Token剪枝 large language model multimodal
9 Visual-Aware CoT: Achieving High-Fidelity Visual Consistency in Unified Models 提出Visual-Aware CoT，提升统一模型在多模态生成中视觉一致性 chain-of-thought
10 CASA: Cross-Attention via Self-Attention for Efficient Vision-Language Fusion 提出CASA：通过自注意力实现的跨注意力，用于高效的视觉-语言融合 multimodal

🔬 支柱三：空间感知与语义 (Perception & Semantics) (5 篇)

#题目一句话要点标签🔗⭐
11 4D Gaussian Splatting as a Learned Dynamical System EvoGS：将4D高斯溅射重构为可学习的动态系统，实现时序一致性动态场景建模 gaussian splatting splatting
12 GaussianImage++: Boosted Image Representation and Compression with 2D Gaussian Splatting GaussianImage++：利用2D高斯溅射增强图像表示与压缩性能 gaussian splatting splatting
13 Retrieving Objects from 3D Scenes with Box-Guided Open-Vocabulary Instance Segmentation 提出基于2D框引导的开放词汇实例分割方法，用于从3D场景中检索目标 open-vocabulary open vocabulary ✅
14 CETCAM: Camera-Controllable Video Generation via Consistent and Extensible Tokenization CETCAM：通过一致且可扩展的Token化实现相机可控的视频生成 depth estimation VGGT geometric consistency ✅
15 WorldWarp: Propagating 3D Geometry with Asynchronous Video Diffusion WorldWarp：利用异步视频扩散传播3D几何信息，生成长时几何一致性视频。 3DGS gaussian splatting splatting ✅

🔬 支柱二：RL算法与架构 (RL & Architecture) (4 篇)

#题目一句话要点标签🔗⭐
16 Anatomy-R1: Enhancing Anatomy Reasoning in Multimodal Large Language Models via Anatomical Similarity Curriculum and Group Diversity Augmentation Anatomy-R1：通过解剖相似性课程学习和群体多样性增强提升多模态大语言模型中的解剖推理能力 curriculum learning large language model multimodal ✅
17 Pushing the Frontier of Audiovisual Perception with Large-Scale Multimodal Correspondence Learning 提出PE-AV：基于大规模对比学习的音视频感知统一编码器，实现跨模态对齐与检索。 contrastive learning multimodal
18 FusionNet: Physics-Aware Representation Learning for Multi-Spectral and Thermal Data via Trainable Signal-Processing Priors FusionNet：通过可训练信号处理先验实现多光谱与热数据的物理感知表征学习 representation learning
19 WaTeRFlow: Watermark Temporal Robustness via Flow Consistency 提出WaTeRFlow框架，增强水印在图像转视频中的时间鲁棒性 world model optical flow

🔬 支柱一：机器人控制 (Robot Control) (3 篇)

#题目一句话要点标签🔗⭐
20 Decoupled Generative Modeling for Human-Object Interaction Synthesis 提出DecHOI，解耦路径规划与动作生成，实现逼真的人-物交互合成 manipulation penetration human-object interaction
21 Zero-shot Reconstruction of In-Scene Object Manipulation from Video 提出首个系统，从单目视频零样本重建场景内物体操作过程。 manipulation scene reconstruction physically plausible
22 VLNVerse: A Benchmark for Vision-Language Navigation with Versatile, Embodied, Realistic Simulation and Evaluation VLNVerse：用于视觉-语言导航的多功能、具身、逼真模拟与评估基准 locomotion sim-to-real embodied AI

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

#题目一句话要点标签🔗⭐
23 PEDESTRIAN: An Egocentric Vision Dataset for Obstacle Detection on Pavements 提出行人视角障碍物检测数据集PEDESTRIAN，用于提升城市人行道安全。 egocentric egocentric vision
24 Hand-Aware Egocentric Motion Reconstruction with Sequence-Level Context 提出HaMoS：一种手部感知的序列级自中心运动重建扩散框架 egocentric egocentric vision

🔬 支柱四：生成式动作 (Generative Motion) (1 篇)

#题目一句话要点标签🔗⭐
25 OmniMoGen: Unifying Human Motion Generation via Learning from Interleaved Text-Motion Instructions OmniMoGen：通过学习交错的文本-动作指令，统一了人体运动生成任务。 text-to-motion motion generation VQ-VAE ✅

🔬 支柱八：物理动画 (Physics-based Animation) (1 篇)

#题目一句话要点标签🔗⭐
26 Towards AI-Guided Open-World Ecological Taxonomic Classification 提出TaxoNet，解决开放世界生态分类中的长尾分布和领域偏移问题。 spatiotemporal foundation model multimodal

⬅️ 返回 cs.CV 首页 · 🏠 返回主页