ICLR 2026 目标检测（object detection）方向上接收论文总结

ICLR 2026目录ICLR 2026通用目标检测框架与方法实时/YOLO/高效检测与 DETR开放词汇/开放世界检测少样本/增量/源域无关检测弱监督/显著性/低光与定向检测3D 目标检测遥感/SAR 与专用场景其他总结参考资料ICLR 2026International Conference on Learning Representations于2026 年 4 月 23 日至 27 日在巴西里约热内卢举行。本文基于 ICLR 2026 Virtual Papers 主会论文列表汇总标题或内容与**目标检测Object Detection**明确相关的论文。目标检测旨在定位图像/点云中的目标并识别其类别。ICLR 更偏重表示学习与算法原理本届 OD 相关工作在DETR/实时检测器、开放词汇、源域无关与增量学习、3D/多模态 BEV、弱监督旋转框与低光/显著性等方向较为集中同时存在大量“Detection”关键词但属于异常/OOD/伪造检测的相邻工作。说明主分类优先依据论文标题中的目标检测相关表述object detection / detector / YOLO / DETR / open-vocabulary detection / 3D object detection 等。标题完全不含目标检测相关表述、或属于异常检测/深度伪造/OOD/图异常等相邻任务的论文归入“其他”。Paper 链接优先 arXiv若检索不到稳定 arXiv 条目则回退到 ICLR Virtual Poster 页面OpenReview forum 链接随会议系统更新个别可能需在 openreview.net 按标题检索。Code/Blog 以公开可检索信息为准未能确认则留空。Team 以 Virtual Poster 作者列表为主单位信息在可确认时补充。现将目标检测方向上接收的论文汇总如下主分类约21篇其他相关 detection 任务见文末。遗漏之处还请大家斧正。通用目标检测框架与方法Unbiased Object Detection Beyond Frequency with Visually Prompted Image SynthesisPaper: https://arxiv.org/abs/2510.18229Code: https://github.com/NUST-Machine-Intelligence-Laboratory/Beyond_FreqKeywords: Unbiased Detection, Class Imbalance, Visual Prompt, Image Synthesis, DebiasingFeatures: 针对检测中的类别频率偏差用视觉提示图像合成做生成式去偏突破仅依赖样本频次重加权的局限Blog:Team: Xinhao Cai; Liulei Li; Gensheng Pei; Tao Chen; Jinshan Pan; Yazhou Yao; Wenguan Wang南京理工大学 Machine Intelligence Laboratory 等Enhancing Vision Transformers for Object Detection via Context-Aware Token Selection and PackingPaper: https://arxiv.org/abs/2410.23608Code:Keywords: Vision Transformer, Token Selection, Token Packing, EfficiencyFeatures: 通过上下文感知的 token 选择与打包增强 ViT 在目标检测上的效率与精度缓解长序列注意力开销Blog:Team: Tianyi Zhang; Baoxin Li; Jae-sun Seo; Yu CaoInlier-Centric Post-Training Quantization for Object Detection ModelsPaper: https://arxiv.org/abs/2602.03472Code:Keywords: Post-Training Quantization, Inlier-Centric, Detector CompressionFeatures: 面向检测模型的后训练量化以内点/内围样本为中心保持定位相关表征压缩同时稳住检测精度Blog:Team: Minsu Kim; Dongyeun Lee; Jaemyung Yu; jiwan hur; Giseop Kim; Junmo KimOD³: Optimization-free Dataset Distillation for Object DetectionPaper: https://arxiv.org/abs/2506.01942Code: https://github.com/VILA-Lab/OD3Keywords: Dataset Distillation, Optimization-free, Dense PredictionFeatures: 提出无优化optimization-free的检测数据蒸馏 OD³降低大数据集上训练检测器的计算成本Blog:Team: Salwa Al Khatib; Ahmed Elhagry; Shitong Shao; Zhiqiang ShenVILA Lab / MBZUAI 等Towards Reliable Detection of Empty Space: Conditional Marked Point Processes for Object DetectionPaper: https://iclr.cc/virtual/2026/poster/10009982Code:Keywords: Empty Space, Marked Point Process, Uncertainty, Safety-criticalFeatures: 用条件标记点过程建模检测强调对“空闲空间/无目标区域”的可靠判定服务安全关键场景Blog:Team: Tobias Riedlinger; Kira Maag; Hanno Gottschalk实时/YOLO/高效检测与 DETRRF-DETR: Neural Architecture Search for Real-Time Detection TransformersPaper: https://arxiv.org/abs/2511.09554Code: https://github.com/roboflow/rf-detrKeywords: RF-DETR, Neural Architecture Search, Real-Time DETR, Open-VocabularyFeatures: 对实时 Detection Transformer 做神经架构搜索RF-DETR面向真实域类别与部署效率开源实现活跃Blog:Team: Isaac Robinson; Peter Robicheaux; Matvei Popov; Deva Ramanan; Neehar PeriRoboflow / CMU 等DiffuDETR: Rethinking Detection Transformers with Denoising Diffusion ProcessPaper: https://iclr.cc/virtual/2026/poster/10007459Code: https://github.com/MBadran2000/DiffuDETRKeywords: DiffuDETR, Denoising Diffusion, Object Query, DETRFeatures: 将目标检测重写为条件 object query 生成结合 DETR 与去噪扩散训练以生成查询/参考点Blog:Team: Youssof Nawar; Mohamed Badran; Marwan TorkiDETR-ViP: Detection Transformer with Robust Discriminative Visual PromptsPaper: https://arxiv.org/abs/2604.14684Code:Keywords: DETR-ViP, Visual Prompt, Discriminative Prompt, Open-VocabularyFeatures: 为 DETR 引入鲁棒可判别视觉提示提升交互式/开放词汇检测中稀有类识别Blog:Team: Bo Qian; Dahu Shi; Xing WeiCertifying the Full YOLO Pipeline: A Probabilistic Verification ApproachPaper: https://iclr.cc/virtual/2026/poster/10011044Code:Keywords: YOLO, Probabilistic Verification, Object Disappearance, SafetyFeatures: 对完整 YOLO 管线做概率化形式验证关注小扰动下目标消失object disappearance风险Blog:Team: Zongxin Liu; Lijia Yu; Tao Lin; Zhiming Chi; Lijun Zhang开放词汇/开放世界检测DeCo-DETR: Decoupled Cognition DETR for efficient Open-Vocabulary Object DetectionPaper: https://arxiv.org/abs/2604.02753Code:Keywords: DeCo-DETR, Open-Vocabulary Detection, Decoupled Cognition, EfficiencyFeatures: 解耦认知式 DETRDeCo-DETR缓解 OV 检测中多模态融合的效率与语义瓶颈Blog:Team: siheng wang; Yanshu Li; Bohan Hu; Zhengdao Li; HaiboZhan; Linshan Li; Weiming Liu; Ruizhi Qian; Guangxin Wu; Hao Zhang; Jifeng Shen; Piotr Koniusz; Zhengtao Yao; Junhao Dong; Qiang Sun多机构合作Retain and Adapt: Auto-Balanced Model Editing for Open-Vocabulary Object Detection under Domain ShiftsPaper: https://iclr.cc/virtual/2026/poster/10011534Code: https://github.com/SCDZX/ICLR26-Auto-Balanced-Model-EditingKeywords: Model Editing, Open-Vocabulary Detection, Domain Shift, Continual AdaptationFeatures: 在域偏移下对 OV 检测器做自动平衡的模型编辑保留旧知识同时适配新域Blog:Team: Zixuan Duan; Fengyuan Lu; Xunzhi Xiang; Wenbin Li; Yang Gao; Qi Fan南京大学 等少样本/增量/源域无关检测FSOD-VFM: Few-Shot Object Detection with Vision Foundation Models and Graph DiffusionPaper: https://arxiv.org/abs/2602.03137Code:Keywords: Few-Shot Object Detection, Vision Foundation Model, Graph DiffusionFeatures: 结合视觉基础模型与图扩散的少样本检测FSOD-VFM强化新类关系建模Blog:Team: Chen-Bin Feng; Youyang Sha; Longfei Liu; Yongjun YU; Chi-Man VONG; Xuanlong Yu; Xi SHENInterference-Isolated Elastic Weight Consolidation and Knowledge Calibration for Incremental Object DetectionPaper: https://iclr.cc/virtual/2026/poster/10009096Code:Keywords: Incremental Object Detection, EWC, Knowledge Calibration, Catastrophic ForgettingFeatures: 用干扰隔离的弹性权重巩固与知识校准缓解增量检测中的遗忘与干扰Blog:Team: De Cheng; Mingyue Zeng; Zhipeng Xu; Di Xu; Nannan Wang; Xinbo GaoCGSA: Class-Guided Slot-Aware Adaptation for Source-Free Object DetectionPaper: https://arxiv.org/abs/2602.22621Code: https://github.com/Michael-McQueen/CGSAKeywords: Source-Free Domain Adaptation, Slot Attention, Pseudo-label, SF-DAODFeatures: CGSA类别引导、slot 感知的源域无关检测适配无需源数据访问Blog:Team: Boyang Dai; 曾 Fan; Zihao Qi; Meng Lou; Yizhou Yu香港大学 等弱监督/显著性/低光与定向检测SPWOOD: Sparse Partial Weakly-Supervised Oriented Object DetectionPaper: https://arxiv.org/abs/2602.03634Code:Keywords: Weakly-Supervised, Oriented Object Detection, Remote Sensing, Sparse AnnotationFeatures: SPWOOD稀疏部分弱监督旋转框检测面向遥感密集目标与昂贵标注Blog:Team: wei zhang; Xiang Liu; Ningjing Liu; Mingxin Liu; Wei Liao; Chunyan Xu; Xue Yang上海交通大学 等S3OD: Towards Generalizable Salient Object Detection with Synthetic DataPaper: https://arxiv.org/abs/2510.21605Code:Keywords: Salient Object Detection, Synthetic Data, Generalization, DIS/HR-SODFeatures: S3OD大规模合成数据提升显著性目标检测的跨任务泛化Blog:Team: Orest Kupyn; Hirokatsu Kataoka; Christian RupprechtUniversity of Oxford 等Self-Guided Low Light Object Detection FrameworkPaper: https://iclr.cc/virtual/2026/poster/10009960Code: https://github.com/gw-shin/SGLDetKeywords: Low-Light Object Detection, Self-Guided, Noise, ContrastFeatures: 自引导低光目标检测框架针对低对比与强噪声下的特征退化Blog:Team: Gwangik Shin; Jaeha Song; Soonmin Hwang3D 目标检测AsyncBEV: Cross-modal flow alignment in Asynchronous 3D Object DetectionPaper: https://arxiv.org/abs/2601.12994Code: https://github.com/tudelft-iv/AsyncBEVKeywords: AsyncBEV, Multi-modal, Asynchronous Sensors, 3D Detection, Flow AlignmentFeatures: AsyncBEV跨模态流对齐处理训练/推理中传感器不同步提升异步设定下的 3D 检测Blog:Team: Shiming Wang; Holger Caesar; Liangliang Nan; Julian KooijTU Delft 等Fore-Mamba3D: Mamba-based Foreground-Enhanced Encoding for 3D Object DetectionPaper: https://arxiv.org/abs/2602.19536Code:Keywords: Fore-Mamba3D, Mamba, Foreground Enhancement, LiDAR, 3D DetectionFeatures: Mamba 前景增强编码抑制体素序列中无用背景提升 3D 检测效率与精度Blog:Team: Zhiwei Ning; Xuanang Gao; Jiaxi Cao; Runze Yang; Huiying Xu; Xinzhong Zhu; Jie Yang; Wei Liu上海交通大学 / 浙江师范大学 等PTNET: A PROPOSAL-CENTRIC TRANSFORMER NETWORK FOR 3D OBJECT DETECTIONPaper: https://iclr.cc/virtual/2026/poster/10008811Code:Keywords: PTNet, Proposal-Centric Transformer, LiDAR, Two-stage 3D DetectionFeatures: 以 proposal 为中心的 Transformer 网络缓解两阶段 3D 检测中几何细节退化Blog:Team: Jianping Zhong; Zhaobo Qi; Kaiwen Duan; Xinyan Liu; Beichen Zhang; Weigang Zhang; Qingming Huang遥感/SAR 与专用场景DNOD: Deformable Neural Operators for Object Detection in SAR ImagesPaper: https://iclr.cc/virtual/2026/poster/10014070Code: https://github.com/quest-lab-iisc/DNODKeywords: SAR, Deformable Neural Operator, Remote Sensing, DETRFeatures: DNOD可变形神经算子用于 SAR 图像目标检测适配雷达散射特性与几何形变Blog:Team: Venkata Sai Mothish Gonugunta; Rishi J; Shobhit Kumar Shukla; Deepak N. SubramaniIISc 等其他以下论文标题含 detection / detector 等关键词但**不属于经典目标检测object detection**主线或任务为异常检测、深度伪造检测、OOD/离群检测、AI 生成内容检测等相邻方向按规则归入“其他”。链接统一指向 ICLR Virtual Poster若有公开 arXiv 可自行按标题补充。All Patches Matter, More Patches Better: Enhance AI-Generated Image Detection via Panoptic Patch LearningPaper: https://iclr.cc/virtual/2026/poster/10007395Code:Keywords: Adjacent Detection TaskFeatures: 面向深度伪造/AI 生成图像或视频的检测与可解释/泛化研究相邻 detection 任务非经典目标检测Blog:Team:A Rich Knowledge Space for Scalable Deepfake DetectionPaper: https://iclr.cc/virtual/2026/poster/10008071Code:Keywords: Adjacent Detection TaskFeatures: 面向深度伪造/AI 生成图像或视频的检测与可解释/泛化研究相邻 detection 任务非经典目标检测Blog:Team:EigenScore: OOD Detection using Posterior Covariance in Diffusion ModelsPaper: https://iclr.cc/virtual/2026/poster/10010743Code:Keywords: Adjacent Detection TaskFeatures: 分布外/离群检测关注模型对未知分布输入的识别而非框级目标定位Blog:Team:Language-guided Open-world Video Anomaly Detection under Weak SupervisionPaper: https://iclr.cc/virtual/2026/poster/10011338Code:Keywords: Adjacent Detection TaskFeatures: 异常检测相关工作与目标检测在“检出异常区域/样本”上相邻但任务设定不同Blog:Team:No Pixel Left Behind: A Detail-Preserving Architecture for Robust High-Resolution AI-Generated Image DetectionPaper: https://iclr.cc/virtual/2026/poster/10011114Code:Keywords: Adjacent Detection TaskFeatures: 面向深度伪造/AI 生成图像或视频的检测与可解释/泛化研究相邻 detection 任务非经典目标检测Blog:Team:VidGuard-R1: AI-Generated Video Detection and Explanation via Reasoning MLLMs and RLPaper: https://iclr.cc/virtual/2026/poster/10008143Code:Keywords: Adjacent Detection TaskFeatures: 面向深度伪造/AI 生成图像或视频的检测与可解释/泛化研究相邻 detection 任务非经典目标检测Blog:Team:FakeXplain: AI-Generated Image Detection via Human-Aligned Grounded ReasoningPaper: https://iclr.cc/virtual/2026/poster/10009205Code:Keywords: Adjacent Detection TaskFeatures: 面向深度伪造/AI 生成图像或视频的检测与可解释/泛化研究相邻 detection 任务非经典目标检测Blog:Team:Veritas: Generalizable Deepfake Detection via Pattern-Aware ReasoningPaper: https://iclr.cc/virtual/2026/poster/10011453Code:Keywords: Adjacent Detection TaskFeatures: 面向深度伪造/AI 生成图像或视频的检测与可解释/泛化研究相邻 detection 任务非经典目标检测Blog:Team:Steering and Rectifying Latent Representation Manifolds in Frozen Multi-modal LLMs for Video Anomaly DetectionPaper: https://iclr.cc/virtual/2026/poster/10007378Code:Keywords: Adjacent Detection TaskFeatures: 异常检测相关工作与目标检测在“检出异常区域/样本”上相邻但任务设定不同Blog:Team:Unveiling Perceptual Artifacts: A Fine-Grained Benchmark for Interpretable AI-Generated Image DetectionPaper: https://iclr.cc/virtual/2026/poster/10009289Code:Keywords: Adjacent Detection TaskFeatures: 面向深度伪造/AI 生成图像或视频的检测与可解释/泛化研究相邻 detection 任务非经典目标检测Blog:Team:Exploring Specular Reflection Inconsistency for Generalizable Face Forgery DetectionPaper: https://iclr.cc/virtual/2026/poster/10010079Code:Keywords: Adjacent Detection TaskFeatures: 面向深度伪造/AI 生成图像或视频的检测与可解释/泛化研究相邻 detection 任务非经典目标检测Blog:Team:Semantic Visual Anomaly Detection and Reasoning in AI-Generated ImagesPaper: https://iclr.cc/virtual/2026/poster/10011907Code:Keywords: Adjacent Detection TaskFeatures: 面向深度伪造/AI 生成图像或视频的检测与可解释/泛化研究相邻 detection 任务非经典目标检测Blog:Team:Preserving Forgery Artifacts: AI-Generated Video Detection at Native ScalePaper: https://iclr.cc/virtual/2026/poster/10008992Code:Keywords: Adjacent Detection TaskFeatures: 面向深度伪造/AI 生成图像或视频的检测与可解释/泛化研究相邻 detection 任务非经典目标检测Blog:Team:Beyond Penalization: Diffusion-based Out-of-Distribution Detection and Selective Regularization in Offline Reinforcement LearningPaper: https://iclr.cc/virtual/2026/poster/10008747Code:Keywords: Adjacent Detection TaskFeatures: 分布外/离群检测关注模型对未知分布输入的识别而非框级目标定位Blog:Team:MRAD: Zero-Shot Anomaly Detection with Memory-Driven RetrievalPaper: https://iclr.cc/virtual/2026/poster/10009313Code:Keywords: Adjacent Detection TaskFeatures: 异常检测相关工作与目标检测在“检出异常区域/样本”上相邻但任务设定不同Blog:Team:Dual Distillation for Few-Shot Anomaly DetectionPaper: https://iclr.cc/virtual/2026/poster/10006980Code:Keywords: Adjacent Detection TaskFeatures: 异常检测相关工作与目标检测在“检出异常区域/样本”上相邻但任务设定不同Blog:Team:总结从本届 ICLR 2026 目标检测相关论文来看呈现以下趋势DETR 家族持续演化RF-DETR实时 NAS、DiffuDETR扩散式 query、DeCo-DETR / DETR-ViP开放词汇与视觉提示表明 Transformer 检测器仍是算法主战场重点转向效率、提示机制与开放类别。数据与学习范式创新OD³ 数据蒸馏、Beyond Frequency 生成式去偏、S3OD 合成显著性数据、SPWOOD 弱监督旋转框强调“少标/合成/蒸馏”降低密集预测成本。适应与可靠部署源域无关 CGSA、增量检测抗遗忘、低光自引导检测、YOLO 管线概率验证、空闲空间可靠估计以及检测器 PTQ体现从精度指标走向域偏移与安全关键可靠性。3D 与非常规传感器AsyncBEV 异步多模态、Fore-Mamba3D / PTNet 点云三维检测、DNOD SAR 神经算子呼应自动驾驶与遥感应用。“Detection” 语义外溢大量异常检测、深度伪造/AIGC 检测、OOD 检测论文与目标检测共享关键词本汇总严格按标题是否体现 object detection 划分主分类与“其他”。总体而言ICLR 2026 的目标检测研究更强调表示与训练范式、开放词汇与适配、以及可部署的可靠/高效检测与 CV 应用会议中的场景密集型工作形成互补。参考资料ICLR 2026 Official WebsiteICLR 2026 Virtual PapersOpenReview注文档基于 ICLR 2026 Virtual Papers 公开列表整理并由 AI 辅助生成arXiv 链接经标题检索核对后写入未能稳定匹配的条目使用 Virtual Poster 链接。Code/Blog 以公开检索结果为准欢迎补充。