cs.CV（2025-09-10）

📊 共 26 篇论文 | 🔗 8 篇有代码

🎯 兴趣领域导航

支柱九：具身大模型 (Embodied Foundation Models) (9 🔗2) 支柱二：RL算法与架构 (RL & Architecture) (8 🔗2) 支柱三：空间感知与语义 (Perception & Semantics) (4 🔗1) 支柱六：视频提取与匹配 (Video Extraction) (2 🔗1) 支柱一：机器人控制 (Robot Control) (2 🔗1) 支柱四：生成式动作 (Generative Motion) (1 🔗1)

🔬 支柱九：具身大模型 (Embodied Foundation Models) (9 篇)

#题目一句话要点标签🔗⭐
1 COCO-Urdu: A Large-Scale Urdu Image-Caption Dataset with Multimodal Quality Estimation COCO-Urdu：构建大规模乌尔都语图像描述数据集，促进多模态研究 large language model multimodal visual grounding
2 Calibrating MLLM-as-a-judge via Multimodal Bayesian Prompt Ensembles 提出MMB方法，通过多模态贝叶斯提示集成校准MLLM在文图生成评判中的偏差。 large language model multimodal
3 Vision-Language Semantic Aggregation Leveraging Foundation Model for Generalizable Medical Image Segmentation 提出基于EM聚合和文本引导解码的医学图像分割方法，提升模型泛化性 foundation model multimodal
4 MITS: A Large-Scale Multimodal Benchmark Dataset for Intelligent Traffic Surveillance 提出MITS大规模多模态数据集，提升智能交通监控领域LMM性能 multimodal instruction following
5 An Open Benchmark Dataset for GeoAI Foundation Models for Oil Palm Mapping in Indonesia 发布印尼油棕种植区GeoAI基础模型开放基准数据集，助力森林砍伐监测。 foundation model PaLM-E
6 Recurrence Meets Transformers for Universal Multimodal Retrieval 提出ReT-2以解决多模态检索任务的复杂性问题 multimodal ✅
7 Retrieval-Augmented VLMs for Multimodal Melanoma Diagnosis 提出检索增强的视觉语言模型，用于提升多模态黑色素瘤诊断的准确性。 multimodal
8 BcQLM: Efficient Vision-Language Understanding with Distilled Q-Gated Cross-Modal Fusion 提出BcQLM：一种轻量级、高效的视觉-语言理解框架，适用于资源受限环境。 large language model multimodal ✅
9 AdsQA: Towards Advertisement Video Understanding 提出AdsQA广告视频问答基准，并设计ReAd-R模型提升LLM在广告理解上的能力。 large language model

🔬 支柱二：RL算法与架构 (RL & Architecture) (8 篇)

#题目一句话要点标签🔗⭐
10 Bitrate-Controlled Diffusion for Disentangling Motion and Content in Video 提出一种基于码率控制扩散模型的视频解耦框架，用于分离视频中的运动和内容。 representation learning motion generation
11 PromptGuard: An Orchestrated Prompting Framework for Principled Synthetic Text Generation for Vulnerable Populations using LLMs with Enhanced Safety, Fairness, and Controllability PromptGuard：针对弱势群体，通过编排式Prompting框架提升LLM生成文本的安全性、公平性和可控性 contrastive learning large language model chain-of-thought
12 First-order State Space Model for Lightweight Image Super-resolution 提出一阶状态空间模型(FSSM)，提升轻量级图像超分辨率性能 Mamba SSM state space model
13 SimCroP: Radiograph Representation Learning with Similarity-driven Cross-granularity Pre-training SimCroP：基于相似性驱动的跨粒度预训练提升胸部CT影像表征学习 representation learning multimodal ✅
14 World Modeling with Probabilistic Structure Integration 提出概率结构集成（PSI），用于学习可控且灵活提示的世界模型。 world model optical flow
15 RewardDance: Reward Scaling in Visual Generation RewardDance：通过生成式奖励建模解决视觉生成中的奖励缩放和奖励利用问题 reinforcement learning RLHF chain-of-thought
16 Hyperspectral Mamba for Hyperspectral Object Tracking 提出基于Mamba的HyMamba网络，用于高光谱目标跟踪，提升复杂场景下的跟踪精度。 Mamba SSM ✅
17 Chirality in Action: Time-Aware Video Representation Learning by Latent Straightening 提出基于潜在空间矫正的时间感知视频表征学习方法，用于手性动作识别。 representation learning

🔬 支柱三：空间感知与语义 (Perception & Semantics) (4 篇)

#题目一句话要点标签🔗⭐
18 Prompt-Driven Image Analysis with Multimodal Generative AI: Detection, Segmentation, Inpainting, and Interpretation 提出基于提示的多模态生成AI图像分析流程，实现检测、分割、修复与描述 open-vocabulary open vocabulary multimodal
19 FractalPINN-Flow: A Fractal-Inspired Network for Unsupervised Optical Flow Estimation with Total Variation Regularization 提出FractalPINN-Flow，一种基于分形网络的无监督光流估计方法。 optical flow
20 UltrON: Ultrasound Occupancy Networks UltrON：利用声学特征的超声图像占据网络，解决弱监督下的三维重建问题 implicit representation geometric consistency ✅
21 Semantic Causality-Aware Vision-Based 3D Occupancy Prediction 提出语义因果感知方法以解决3D占用预测问题 semantic mapping semantic map

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

#题目一句话要点标签🔗⭐
22 iMatcher: Improve matching in point cloud registration via local-to-global geometric consistency learning iMatcher：通过局部到全局几何一致性学习改进点云配准中的特征匹配 feature matching geometric consistency
23 Diffusion-Based Action Recognition Generalizes to Untrained Domains 提出基于扩散模型的动作识别方法，提升模型在未训练域上的泛化能力 egocentric ✅

🔬 支柱一：机器人控制 (Robot Control) (2 篇)

#题目一句话要点标签🔗⭐
24 EfficientIML: Efficient High-Resolution Image Manipulation Localization 提出EfficientIML模型，用于高效高分辨率图像篡改定位，并构建了高分辨率SIF数据集。 manipulation
25 ArgoTweak: Towards Self-Updating HD Maps through Structured Priors ArgoTweak：通过结构化先验实现高精地图的自更新 sim2real ✅

🔬 支柱四：生成式动作 (Generative Motion) (1 篇)

#题目一句话要点标签🔗⭐
26 HuMo: Human-Centric Video Generation via Collaborative Multi-Modal Conditioning HuMo：通过协同多模态条件控制实现以人为中心的视频生成 classifier-free guidance foundation model multimodal ✅

⬅️ 返回 cs.CV 首页 · 🏠 返回主页