ICCV 2025 目标检测（object detection）方向上接收论文总结

yyywxk2026-08-25 8:36

ICCV 2025

目录

  • [ICCV 2025](#ICCV 2025)
    • [开放词汇 / 开放集 / 图像提示检测](#开放词汇 / 开放集 / 图像提示检测)
    • [3D 目标检测](#3D 目标检测)
    • 伪装目标检测
    • [小目标 / 遥感 / 航拍 / 在轨检测](#小目标 / 遥感 / 航拍 / 在轨检测)
    • [域自适应 / 域泛化 / 测试时适应](#域自适应 / 域泛化 / 测试时适应)
    • [多模态 / 事件 / RAW / 低光 / 雷达检测](#多模态 / 事件 / RAW / 低光 / 雷达检测)
    • [半监督 / 少样本 / 增量 / 无标注学习](#半监督 / 少样本 / 增量 / 无标注学习)
    • [鲁棒性 / 对抗攻击与防御](#鲁棒性 / 对抗攻击与防御)
    • 人-物交互检测（HOI）
    • 其他：评测、生成、推理与量化
  • 总结
  • 参考资料

ICCV 2025（International Conference on Computer Vision）于 2025 年 10 月 19 日至 23 日在美国夏威夷檀香山举行。

目标检测（Object Detection）旨在定位并识别图像、视频或点云中的感兴趣目标，是自动驾驶、遥感解译、安防监控和机器人感知等应用的基础能力。

现根据 ICCV 2025 Open Access 的公开列表，以题名明确包含目标检测、检测器、3D Detection、红外小目标检测或人--物交互检测等表述为主线，汇总相关接收论文 82 篇。为避免把显著性检测、异常检测、语言定位等同名但任务定义不同的方向混入，未采用仅凭摘要泛化扩展的口径。

链接说明：Paper 优先列出已核对题名的 arXiv 链接，无 arXiv 时回退到 CVF Open Access；Code 仅列出确认与论文对应的公开仓库；Blog 仅列单篇精讲，未找到可靠精讲时不显示该字段；Team 以论文署名机构为主，个别条目按公开作者/项目页补充。

开放词汇 / 开放集 / 图像提示检测

  1. 3D-MOOD: Lifting 2D to 3D for Monocular Open-Set Object Detection
  1. ASGS: Single-Domain Generalizable Open-Set Object Detection via Adaptive Subgraph Searching
  1. Benefit From Seen: Enhancing Open-Vocabulary Object Detection by Bridging Visual and Textual Co-Occurrence Knowledge
  1. Dynamic-DINO: Fine-Grained Mixture of Experts Tuning for Real-time Open-Vocabulary Object Detection
  1. OpenM3D: Open Vocabulary Multi-view Indoor 3D Object Detection without Human Annotations
  • Paper: https://arxiv.org/abs/2508.20063
  • Keywords: Open Vocabulary, Multi-view Indoor 3D Detection, No Annotations
  • Features: 无需人工标注的开放词汇多视角室内3D目标检测
  • Team: National Yang Ming Chiao Tung University
  1. OpenRSD: Towards Open-prompts for Object Detection in Remote Sensing Images
  • Paper: https://arxiv.org/abs/2503.06146
  • Blog: 论文解读：OpenRSD
  • Keywords: Remote Sensing, Open-Prompts, Object Detection
  • Features: 支持文本和图像开放提示、水平框与旋转框检测，并以大规模 ORSD+ 数据完成三阶段训练
  • Team: 北京航空航天大学
  1. OV-SCAN: Semantically Consistent Alignment for Novel Object Discovery in Open-Vocabulary 3D Object Detection
  • Paper: https://arxiv.org/abs/2503.06435
  • Keywords: Open-Vocabulary 3D Detection, Novel Object Discovery, Semantic Alignment
  • Features: 语义一致性对齐用于开放词汇3D目标检测中的新目标发现
  • Team: University of Waterloo
  1. SFUOD: Source-Free Unknown Object Detection
  1. Superpowering Open-Vocabulary Object Detectors for X-ray Vision
  • Paper: https://arxiv.org/abs/2503.17071
  • Code: https://github.com/PAGF188/RAXO
  • Keywords: Open-Vocabulary Detection, X-ray Vision, Superpowering
  • Features: 以训练免调的 RAXO 将 RGB 开放词汇检测器适配到 X 光，并发布 300 余类的 DET-COMPASS 基准
  • Team: University of Santiago de Compostela；Fondazione Bruno Kessler；University of Trento
  1. Visual Modality Prompt for Adapting Vision-Language Object Detectors
  1. Visual Textualization for Image Prompted Object Detection

3D 目标检测

  1. Accelerate 3D Object Detection Models via Zero-Shot Attention Key Pruning
  • Paper: https://arxiv.org/abs/2503.08101
  • Keywords: 3D Object Detection, Attention Pruning, Zero-Shot, Model Acceleration
  • Features: 零样本注意力键剪枝加速3D目标检测模型
  • Team: 北京交通大学
  1. Adaptive Dual Uncertainty Optimization: Boosting Monocular 3D Object Detection under Test-Time Shifts
  • Paper: https://arxiv.org/abs/2508.20488
  • Keywords: Monocular 3D Object Detection, Test-Time Adaptation, Uncertainty Optimization
  • Features: 自适应双不确定性优化提升测试时偏移下的单目3D检测
  • Team: 北京大学
  1. Boosting Multi-View Indoor 3D Object Detection via Adaptive 3D Volume Construction
  • Paper: https://arxiv.org/abs/2507.18331
  • Keywords: Multi-View Indoor, 3D Object Detection, Adaptive Volume
  • Features: 自适应3D体积构建增强多视角室内3D目标检测
  • Team: 浙江大学
  1. CHARM3R: Towards Unseen Camera Height Robust Monocular 3D Detector
  • Paper: https://arxiv.org/abs/2508.11185
  • Keywords: Monocular 3D Detection, Camera Height Robustness
  • Features: 面向未见相机高度鲁棒的单目3D检测器
  • Team: Michigan State University / General Motors
  1. CVFusion: Cross-View Fusion of 4D Radar and Camera for 3D Object Detection
  • Paper: https://arxiv.org/abs/2507.04587
  • Keywords: 4D Radar, Cross-View Fusion, 3D Object Detection
  • Features: 4D雷达与相机跨视角融合用于3D目标检测
  • Team: 厦门大学
  1. DiffRefine: Diffusion-based Proposal Specific Point Cloud Densification for Cross-Domain Object Detection
  1. Doppler-Aware LiDAR-RADAR Fusion for Weather-Robust 3D Detection
  1. EVT: Efficient View Transformation for Multi-Modal 3D Object Detection
  • Paper: https://arxiv.org/abs/2411.10715
  • Keywords: View Transformation, Multi-Modal 3D Detection, Efficient
  • Features: 高效视角变换用于多模态3D目标检测
  • Team: ETRI / KAIST
  1. FreqPDE: Rethinking Positional Depth Embedding for Multi-View 3D Object Detection Transformers
  • Paper: https://arxiv.org/abs/2510.15385
  • Keywords: Multi-View 3D Detection, Positional Depth Embedding, Transformer
  • Features: 重新思考多视角3D目标检测Transformer中的位置深度嵌入
  • Team: 西安交通大学
  1. GeoFormer: Geometry Point Encoder for 3D Object Detection with Graph-based Transformer
  1. Harnessing Uncertainty-aware Bounding Boxes for Unsupervised 3D Object Detection
  • Paper: https://arxiv.org/abs/2408.00619
  • Keywords: Unsupervised 3D Detection, Uncertainty-aware Bounding Boxes
  • Features: 利用不确定性感知边界框进行无监督3D目标检测
  • Team: 复旦大学
  1. Height-Fidelity Dense Global Fusion for Multi-modal 3D Object Detection
  • Paper: https://arxiv.org/abs/2507.04369
  • Code: https://github.com/AutoLab-SAI-SJTU/MambaFusion
  • Keywords: Multi-modal 3D Detection, Height-Fidelity, Dense Global Fusion
  • Features: 以高度保真的 LiDAR 编码和 Hybrid Mamba 实现高效的相机--LiDAR 密集全局融合
  • Team: 中国科学院自动化研究所；中国科学院大学；上海交通大学；Anyverse Intelligence；上海科技大学
  1. MemDistill: Distilling LiDAR Knowledge into Memory for Camera-Only 3D Object Detection
  1. MonoSOWA: Scalable Monocular 3D Object Detector Without Human Annotations
  1. Motal: Unsupervised 3D Object Detection by Modality and Task-specific Knowledge Transfer
  1. OcRFDet: Object-Centric Radiance Fields for Multi-View 3D Object Detection in Autonomous Driving
  1. Perspective-Invariant 3D Object Detection
  • Paper: https://arxiv.org/abs/2507.17665
  • Keywords: Perspective-Invariant, 3D Object Detection
  • Features: 视角不变的3D目标检测方法
  • Team: National University of Singapore
  1. Power of Cooperative Supervision: Multiple Teachers Framework for Advanced 3D Semi-Supervised Object Detection
  1. RCTDistill: Cross-Modal Knowledge Distillation Framework for Radar-Camera 3D Object Detection with Temporal Fusion
  • Paper: https://arxiv.org/abs/2509.17712
  • Keywords: Radar-Camera 3D Detection, Knowledge Distillation, Temporal Fusion
  • Features: 将雷达--相机跨模态知识蒸馏与时序融合结合，用于 3D 目标检测
  • Team: Yonsei University
  1. Robust 3D Object Detection using Probabilistic Point Clouds from Single-Photon LiDARs
  1. Towards Accurate and Efficient 3D Object Detection for Autonomous Driving: A Mixture of Experts Computing System on Edge
  1. Unleashing the Temporal Potential of Stereo Event Cameras for Continuous-Time 3D Object Detection

伪装目标检测

  1. Beyond Single Images: Retrieval Self-Augmented Unsupervised Camouflaged Object Detection
  • Paper: https://arxiv.org/abs/2510.18437
  • Keywords: Camouflaged Object Detection, Retrieval Self-Augmented, Unsupervised
  • Features: 基于检索自增强的无监督伪装目标检测
  • Team: 华南理工大学
  1. Enhancing Prompt Generation with Adaptive Refinement for Camouflaged Object Detection
  1. ESCNet:Edge-Semantic Collaborative Network for Camouflaged Object Detection
  1. Improving SAM for Camouflaged Object Detection via Dual Stream Adapters
  • Paper: https://arxiv.org/abs/2503.06042
  • Keywords: SAM, Camouflaged Object Detection, Dual Stream Adapters
  • Features: 用双流适配器增强 SAM 在伪装目标检测中的特征提取与掩码生成
  • Team: 上海交通大学

小目标 / 遥感 / 航拍 / 在轨检测

  1. Active Learning Meets Foundation Models: Fast Remote Sensing Data Annotation for Object Detection
  1. Adapting Vehicle Detectors for Aerial Imagery to Unseen Domains with Weak Supervision
  • Paper: https://arxiv.org/abs/2507.20976
  • Keywords: Aerial Detection, Vehicle Detectors, Weak Supervision, Domain Adaptation
  • Features: 弱监督将车辆检测器适配到未见航拍域
  • Team: Carnegie Mellon University
  1. DM-EFS: Dynamically Multiplexed Expanded Features Set Form for Robust and Efficient Small Object Detection
  1. Dual Domain Control via Active Learning for Remote Sensing Domain Incremental Object Detection
  1. Event-based Tiny Object Detection: A Benchmark Dataset and Baseline
  1. From Easy to Hard: Progressive Active Learning Framework for Infrared Small Target Detection with Single Point Supervision
  1. Fusion Meets Diverse Conditions: A High-diversity Benchmark and Baseline for UAV-based Multimodal Object Detection with Condition Cues
  • Paper: https://arxiv.org/abs/2510.13620
  • Keywords: UAV, Multimodal Object Detection, Diverse Conditions, Benchmark
  • Features: 构建带条件线索的高多样性无人机多模态检测基准与基线
  • Team: 中国电子科技集团
  1. LLM-Assisted Semantic Guidance for Sparsely Annotated Remote Sensing Object Detection
  1. Measuring the Impact of Rotation Equivariance on Aerial Object Detection
  1. Text-IRSTD: Leveraging Semantic Text to Promote Infrared Small Target Detection in Complex Scenes
  1. Uncertainty-Aware Gradient Stabilization for Small Object Detection
  1. VISO: Accelerating In-orbit Object Detection with Language-Guided Mask Learning and Sparse Inference

域自适应 / 域泛化 / 测试时适应

  1. Boosting Domain Generalized and Adaptive Detection with Diffusion Models: Fitness, Generalization, and Transferability
  • Paper: https://arxiv.org/abs/2506.21042
  • Keywords: Domain Generalization, Adaptive Detection, Diffusion Models
  • Features: 利用扩散模型提高域泛化与自适应检测的适配性、泛化性和可迁移性
  • Team: 浙江工业大学
  1. Continual Adaptation: Environment-Conditional Parameter Generation for Object Detection in Dynamic Scenarios
  • Paper: https://arxiv.org/abs/2506.24063
  • Keywords: Continual Adaptation, Parameter Generation, Dynamic Scenarios
  • Features: 环境条件参数生成实现动态场景下的持续自适应目标检测
  • Team: 天津大学
  1. Debiased Teacher for Day-to-Night Domain Adaptive Object Detection
  1. Diffusion-based Source-biased Model for Single Domain Generalized Object Detection
  1. Dual-Rate Dynamic Teacher for Source-Free Domain Adaptive Object Detection
  1. UPRE: Zero-Shot Domain Adaptation for Object Detection via Unified Prompt and Representation Enhancement

多模态 / 事件 / RAW / 低光 / 雷达检测

  1. Beyond RGB: Adaptive Parallel Processing for RAW Object Detection
  • Paper: https://arxiv.org/abs/2503.13163
  • Keywords: RAW Object Detection, Adaptive Parallel Processing
  • Features: 自适应并行处理在RAW域直接进行目标检测
  • Team: Sony
  1. Dark-ISP: Enhancing RAW Image Processing for Low-Light Object Detection
  • Paper: https://arxiv.org/abs/2509.09183
  • Keywords: RAW Image Processing, Low-Light Detection, Dark-ISP
  • Features: 增强RAW图像处理用于低光目标检测
  • Team: 上海交通大学
  1. DoppDrive: Doppler-Driven Temporal Aggregation for Improved Radar Object Detection
  1. EvRT-DETR: Latent Space Adaptation of Image Detectors for Event-based Vision
  1. ForeSight: Multi-View Streaming Joint Object Detection and Trajectory Forecasting
  • Paper: https://arxiv.org/abs/2508.07089
  • Keywords: Multi-View, Joint Detection and Forecasting, Streaming
  • Features: 多视角流式联合目标检测与轨迹预测
  • Team: University of Toronto
  1. Rethinking Multi-modal Object Detection from the Perspective of Mono-Modality Feature Learning
  • Paper: https://arxiv.org/abs/2503.11780
  • Keywords: Multi-modal Detection, Mono-Modality Feature Learning
  • Features: 从单模态特征学习角度重构多模态检测，改善各模态互补信息的学习与融合
  • Team: 北京理工大学
  1. WaveMamba: Wavelet-Driven Mamba Fusion for RGB-Infrared Object Detection
  • Paper: https://arxiv.org/abs/2507.18173
  • Keywords: RGB-Infrared Detection, Wavelet, Mamba Fusion
  • Features: 用离散小波分解和 Mamba 融合 RGB/红外高低频信息，并以逆小波检测头减少信息损失
  • Team: 北京航空航天大学等

半监督 / 少样本 / 增量 / 无标注学习

  1. AnnofreeOD: Detecting All Classes at Low Frame Rates Without Human Annotations
  1. DuET: Dual Incremental Object Detection via Exemplar-Free Task Arithmetic
  • Paper: https://arxiv.org/abs/2506.21260
  • Keywords: Incremental Object Detection, Task Arithmetic, Exemplar-Free
  • Features: 用无样本任务算术同时处理类别增量和域增量检测，避免保存旧类样本
  • Team: IIT Bombay
  1. Gradient Decomposition and Alignment for Incremental Object Detection
  1. STEP-DETR: Advancing DETR-based Semi-Supervised Object Detection with Super Teacher and Pseudo-Label Guided Text Queries
  1. When Pixel Difference Patterns Meet ViT: PiDiViT for Few-Shot Object Detection

鲁棒性 / 对抗攻击与防御

  1. Adversarial Attention Perturbations for Large Object Detection Transformers
  1. Gradient-Reweighted Adversarial Camouflage for Physical Object Detection Evasion
  1. PBCAT: Patch-Based Composite Adversarial Training against Physically Realizable Attacks on Object Detection
  • Paper: https://arxiv.org/abs/2506.23581
  • Keywords: Adversarial Training, Patch Attack, Physical Realizable
  • Features: 基于补丁的复合对抗训练防御物理可实现对目标检测的攻击
  • Team: 清华大学
  1. Revisiting Adversarial Patch Defenses on Object Detectors: Unified Evaluation, Large-Scale Dataset, and New Insights

人-物交互检测（HOI）

  1. Bilateral Collaboration with Large Vision-Language Models for Open Vocabulary Human-Object Interaction Detection
  • Paper: https://arxiv.org/abs/2507.06510
  • Keywords: Open Vocabulary HOI Detection, Vision-Language Models, Bilateral Collaboration
  • Features: 以大型视觉语言模型的双边协作增强开放词汇 HOI 检测
  • Team: 华南理工大学；新加坡科技研究局（A*STAR）
  1. HOLa: Zero-Shot HOI Detection with Low-Rank Decomposed VLM Feature Adaptation
  • Paper: https://arxiv.org/abs/2507.15542
  • Code: https://github.com/ChelsieLei/HOLa
  • Keywords: Open-Vocabulary HOI Detection, Low-Rank Decomposition, VLM Feature Adaptation
  • Features: 低秩分解 VLM 特征并用 LLM 动作正则化，提升零样本 HOI 对未见动作的泛化
  • Team: National University of Singapore；University of Mississippi；ASUS Intelligent Cloud Services
  1. No More Sibling Rivalry: Debiasing Human-Object Interaction Detection
  • Paper: https://arxiv.org/abs/2509.00760
  • Keywords: Human-Object Interaction Detection, Debiasing, Sibling Rivalry
  • Features: 消除人-物交互检测中的兄弟竞争偏差
  • Team: 上海交通大学
  1. Open-Vocabulary HOI Detection with Interaction-aware Prompt and Concept Calibration
  • Paper: https://arxiv.org/abs/2508.03207
  • Keywords: Open-Vocabulary HOI Detection, Interaction-aware Prompt, Concept Calibration
  • Features: 交互感知提示与概念校准实现开放词汇HOI检测
  • Team: 北京大学
  1. Visual Relation Diffusion for Human-Object Interaction Detection

其他：评测、生成、推理与量化

  1. Automated Model Evaluation for Object Detection via Prediction Consistency and Reliability
  1. Cycle-Consistent Learning for Joint Layout-to-Image Generation and Object Detection
  1. From Objects to Events: Unlocking Complex Visual Understanding in Object Detectors via LLM-guided Symbolic Reasoning
  • Paper: https://arxiv.org/abs/2502.05843
  • Keywords: LLM, Symbolic Reasoning, Event Understanding, Object Detectors
  • Features: 用 LLM 引导的符号推理把目标检测器输出扩展为复杂视觉事件理解
  • Team: 北京大学；厦门大学
  1. LMM-Det: Make Large Multimodal Models Excel in Object Detection
  1. Task-Specific Zero-shot Quantization-Aware Training for Object Detection
  • Paper: https://arxiv.org/abs/2507.16782
  • Code: https://github.com/DFQ-Dojo/dfq-toolkit
  • Keywords: Zero-Shot Quantization, Task-Specific Calibration, Knowledge Distillation
  • Features: 无需原始训练数据，通过框与类别采样合成任务特定校准集，再以任务特定蒸馏恢复量化检测器性能
  • Team: Georgia Institute of Technology；清华大学
  1. Unified Category-Level Object Detection and Pose Estimation from RGB Images using 3D Prototypes
  • Paper: https://arxiv.org/abs/2508.02157
  • Keywords: Category-Level, Object Detection, Pose Estimation, 3D Prototypes
  • Features: 使用3D原型统一类别级目标检测与姿态估计
  • Team: Saarland University

总结

从题名明确指向目标检测的 82 篇论文来看，ICCV 2025 呈现出以下趋势：

  1. 3D 感知仍是最大板块：22 篇论文覆盖单目、多视角、相机--LiDAR、雷达--相机、事件相机和单光子 LiDAR，并同时关注无标注、半监督、跨域与边缘部署。

  2. 开放世界能力继续外扩：11 篇开放词汇、开放集或图像提示检测工作，将能力扩展到单目 3D、室内多视角、遥感、X 光和新模态适配。

  3. 遥感、小目标与极端成像条件受到集中关注：12 篇论文面向航拍、遥感、红外小目标、事件相机反无人机与在轨检测，主动学习、弱监督和旋转等变是高频技术路线。

  4. 真实部署推动多模态和效率设计：RAW/低光、RGB--红外、雷达、事件流以及流式检测等工作强调传感器互补；MoE、Mamba、剪枝、量化和稀疏推理则面向实时或边缘场景。

  5. 数据与标注约束成为核心问题：无人工标注、稀疏标注、半监督、增量学习和无源域适应贯穿多个分类，说明检测器正从静态闭集训练走向持续、低标注成本的开放环境。

  6. 安全性与高层理解同步推进：物理对抗伪装、对抗补丁防御和 Transformer 攻击反映了部署安全需求；HOI、LLM 符号推理与大多模态检测则推动检测结果向关系和事件理解延伸。

参考资料

  1. ICCV 2025 Open Access Repository
  2. ICCV 2025 论文精讲索引（AI Paper Notes）

注：文档部分内容由 AI 辅助整理。链接与题名已进行一致性核对，但团队机构采用公开页面汇总口径，仍建议正式引用前以论文 PDF 首页为准。

相关推荐
欧特克_Glodon40 分钟前
OpenCV计算机视觉开发入门与实践＜十七＞：点运算与灰度变换概述
c++·人工智能·opencv·计算机视觉
chen_zn9540 分钟前
《VLA 系列》MemoryVLA++ | 感知-认知记忆 | 潜空间未来想象 | 论文与源码边界解析
人工智能·具身智能·vla
狂云歌40 分钟前
AI时代，学什么，怎么学
人工智能·学习
Allen_LVyingbo42 分钟前
医疗AI可扩展网格信息系统中网格计算技术的智能优化
大数据·人工智能·python·算法·机器学习·django·健康医疗
一个帅气昵称啊42 分钟前
.Net C# AI智能体开发-AI智能体
人工智能·c#·.net
MNLoser43 分钟前
AI agent开发——LangGraph接入持久化
linux·人工智能·windows·python
民乐团扒谱机1 小时前
【读论文】2005 一种旋律转录的分类方法A classification approach to melody transcription[C]
人工智能·分类·数据挖掘
程序员-李俞1 小时前
Coze 工作流调用异步 HTTP API 完整教程：任务 ID、循环轮询、状态判断与结果 URL 提取
网络·人工智能·网络协议·http·aigc·ai编程·ai写作
柠檬07111 小时前
图像亮暗不均匀，如何去除背景或者怎么处理可以让图像暗的地方亮一些，亮的地方暗一些
人工智能·opencv·计算机视觉
热门推荐
01国内可直接用、免费额度/永久免费的大模型API清单（含 SiliconFlow、火山、阿里、智谱、百度、Kimi、DeepSeek、DMXAPI 等）02GitHub 镜像站点03微信历史版本含下载地址（ Windows PC | 安卓 | MAC ）及设置微信不更新042026 国产 AI 大模型横评：DeepSeek、通义千问、Kimi、文心一言、星火、豆包谁更能打？05历年考研数学一、数学二、数学三真题试卷及答案PDF06Agnes AI 免费 API 接入指南：文本、生图、生视频，一套接口全免费07AI 编程 IDE 全景解析 2026：Agent 全面接管开发链路082026年7月AI圈大地震：GPT-5.6被政府限制、Claude入驻Slack、Anthropic自研芯片09VSCode+CodeX扩展，无法启动10如何新建文件夹？ 电脑新建文件夹的4种方法