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

yyywxk2026-09-06 8:45

ICLR 2026

目录

ICLR 2026（International 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：

  1. 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）。
  2. 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"。
  3. 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）。
  4. 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

  1. Unbiased Object Detection Beyond Frequency with Visually Prompted Image Synthesis
  • Paper: https://arxiv.org/abs/2510.18229
  • Code: https://github.com/NUST-Machine-Intelligence-Laboratory/Beyond_Freq
  • Keywords: Unbiased Detection, Class Imbalance, Visual Prompt, Image Synthesis, Debiasing
  • Features: 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）
  1. Enhancing Vision Transformers for Object Detection via Context-Aware Token Selection and Packing
  • Paper: https://arxiv.org/abs/2410.23608
  • Code:
  • Keywords: Vision Transformer, Token Selection, Token Packing, Efficiency
  • Features: 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 Cao
  1. Inlier-Centric Post-Training Quantization for Object Detection Models
  • Paper: https://arxiv.org/abs/2602.03472
  • Code:
  • Keywords: Post-Training Quantization, Inlier-Centric, Detector Compression
  • Features: 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 Kim
  1. OD³: Optimization-free Dataset Distillation for Object Detection
  • Paper: https://arxiv.org/abs/2506.01942
  • Code: https://github.com/VILA-Lab/OD3
  • Keywords: Dataset Distillation, Optimization-free, Dense Prediction
  • Features: 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 Shen（VILA Lab / MBZUAI 等）
  1. Towards Reliable Detection of Empty Space: Conditional Marked Point Processes for Object Detection
  • Paper: https://iclr.cc/virtual/2026/poster/10009982
  • Code:
  • Keywords: Empty Space, Marked Point Process, Uncertainty, Safety-critical
  • Features: 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 DETR

  1. RF-DETR: Neural Architecture Search for Real-Time Detection Transformers
  • Paper: https://arxiv.org/abs/2511.09554
  • Code: https://github.com/roboflow/rf-detr
  • Keywords: RF-DETR, Neural Architecture Search, Real-Time DETR, Open-Vocabulary
  • Features: 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 Peri（Roboflow / CMU 等）
  1. DiffuDETR: Rethinking Detection Transformers with Denoising Diffusion Process
  1. DETR-ViP: Detection Transformer with Robust Discriminative Visual Prompts
  • Paper: https://arxiv.org/abs/2604.14684
  • Code:
  • Keywords: DETR-ViP, Visual Prompt, Discriminative Prompt, Open-Vocabulary
  • Features: 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 Wei
  1. Certifying the Full YOLO Pipeline: A Probabilistic Verification Approach
  • Paper: https://iclr.cc/virtual/2026/poster/10011044
  • Code:
  • Keywords: YOLO, Probabilistic Verification, Object Disappearance, Safety
  • Features: 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

  1. DeCo-DETR: Decoupled Cognition DETR for efficient Open-Vocabulary Object Detection
  • Paper: https://arxiv.org/abs/2604.02753
  • Code:
  • Keywords: DeCo-DETR, Open-Vocabulary Detection, Decoupled Cognition, Efficiency
  • Features: 解耦认知式 DETR（DeCo-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）
  1. Retain and Adapt: Auto-Balanced Model Editing for Open-Vocabulary Object Detection under Domain Shifts

少样本/增量/源域无关检测

  1. FSOD-VFM: Few-Shot Object Detection with Vision Foundation Models and Graph Diffusion
  • Paper: https://arxiv.org/abs/2602.03137
  • Code:
  • Keywords: Few-Shot Object Detection, Vision Foundation Model, Graph Diffusion
  • Features: 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 SHEN
  1. Interference-Isolated Elastic Weight Consolidation and Knowledge Calibration for Incremental Object Detection
  • Paper: https://iclr.cc/virtual/2026/poster/10009096
  • Code:
  • Keywords: Incremental Object Detection, EWC, Knowledge Calibration, Catastrophic Forgetting
  • Features: 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 Gao
  1. CGSA: Class-Guided Slot-Aware Adaptation for Source-Free Object Detection

弱监督/显著性/低光与定向检测

  1. SPWOOD: Sparse Partial Weakly-Supervised Oriented Object Detection
  • Paper: https://arxiv.org/abs/2602.03634
  • Code:
  • Keywords: Weakly-Supervised, Oriented Object Detection, Remote Sensing, Sparse Annotation
  • Features: 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）
  1. S3OD: Towards Generalizable Salient Object Detection with Synthetic Data
  • Paper: https://arxiv.org/abs/2510.21605
  • Code:
  • Keywords: Salient Object Detection, Synthetic Data, Generalization, DIS/HR-SOD
  • Features: 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 Rupprecht（University of Oxford 等）
  1. Self-Guided Low Light Object Detection Framework

3D 目标检测

  1. AsyncBEV: Cross-modal flow alignment in Asynchronous 3D Object Detection
  • Paper: https://arxiv.org/abs/2601.12994
  • Code: https://github.com/tudelft-iv/AsyncBEV
  • Keywords: AsyncBEV, Multi-modal, Asynchronous Sensors, 3D Detection, Flow Alignment
  • Features: 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 Kooij（TU Delft 等）
  1. Fore-Mamba3D: Mamba-based Foreground-Enhanced Encoding for 3D Object Detection
  • Paper: https://arxiv.org/abs/2602.19536
  • Code:
  • Keywords: Fore-Mamba3D, Mamba, Foreground Enhancement, LiDAR, 3D Detection
  • Features: 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）
  1. PTNET: A PROPOSAL-CENTRIC TRANSFORMER NETWORK FOR 3D OBJECT DETECTION
  • Paper: https://iclr.cc/virtual/2026/poster/10008811
  • Code:
  • Keywords: PTNet, Proposal-Centric Transformer, LiDAR, Two-stage 3D Detection
  • Features: 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

  1. DNOD: Deformable Neural Operators for Object Detection in SAR Images

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。

  1. All Patches Matter, More Patches Better: Enhance AI-Generated Image Detection via Panoptic Patch Learning
  • Paper: https://iclr.cc/virtual/2026/poster/10007395
  • Code:
  • Keywords: Adjacent Detection Task
  • Features: 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:
  1. A Rich Knowledge Space for Scalable Deepfake Detection
  • Paper: https://iclr.cc/virtual/2026/poster/10008071
  • Code:
  • Keywords: Adjacent Detection Task
  • Features: 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:
  1. EigenScore: OOD Detection using Posterior Covariance in Diffusion Models
  1. Language-guided Open-world Video Anomaly Detection under Weak Supervision
  • Paper: https://iclr.cc/virtual/2026/poster/10011338
  • Code:
  • Keywords: Adjacent Detection Task
  • Features: 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:
  1. No Pixel Left Behind: A Detail-Preserving Architecture for Robust High-Resolution AI-Generated Image Detection
  • Paper: https://iclr.cc/virtual/2026/poster/10011114
  • Code:
  • Keywords: Adjacent Detection Task
  • Features: 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:
  1. VidGuard-R1: AI-Generated Video Detection and Explanation via Reasoning MLLMs and RL
  • Paper: https://iclr.cc/virtual/2026/poster/10008143
  • Code:
  • Keywords: Adjacent Detection Task
  • Features: 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:
  1. FakeXplain: AI-Generated Image Detection via Human-Aligned Grounded Reasoning
  • Paper: https://iclr.cc/virtual/2026/poster/10009205
  • Code:
  • Keywords: Adjacent Detection Task
  • Features: 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:
  1. Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoning
  • Paper: https://iclr.cc/virtual/2026/poster/10011453
  • Code:
  • Keywords: Adjacent Detection Task
  • Features: 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:
  1. Steering and Rectifying Latent Representation Manifolds in Frozen Multi-modal LLMs for Video Anomaly Detection
  • Paper: https://iclr.cc/virtual/2026/poster/10007378
  • Code:
  • Keywords: Adjacent Detection Task
  • Features: 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:
  1. Unveiling Perceptual Artifacts: A Fine-Grained Benchmark for Interpretable AI-Generated Image Detection
  • Paper: https://iclr.cc/virtual/2026/poster/10009289
  • Code:
  • Keywords: Adjacent Detection Task
  • Features: 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:
  1. Exploring Specular Reflection Inconsistency for Generalizable Face Forgery Detection
  • Paper: https://iclr.cc/virtual/2026/poster/10010079
  • Code:
  • Keywords: Adjacent Detection Task
  • Features: 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:
  1. Semantic Visual Anomaly Detection and Reasoning in AI-Generated Images
  • Paper: https://iclr.cc/virtual/2026/poster/10011907
  • Code:
  • Keywords: Adjacent Detection Task
  • Features: 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:
  1. Preserving Forgery Artifacts: AI-Generated Video Detection at Native Scale
  • Paper: https://iclr.cc/virtual/2026/poster/10008992
  • Code:
  • Keywords: Adjacent Detection Task
  • Features: 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:
  1. Beyond Penalization: Diffusion-based Out-of-Distribution Detection and Selective Regularization in Offline Reinforcement Learning
  1. MRAD: Zero-Shot Anomaly Detection with Memory-Driven Retrieval
  • Paper: https://iclr.cc/virtual/2026/poster/10009313
  • Code:
  • Keywords: Adjacent Detection Task
  • Features: 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:
  1. Dual Distillation for Few-Shot Anomaly Detection
  • Paper: https://iclr.cc/virtual/2026/poster/10006980
  • Code:
  • Keywords: Adjacent Detection Task
  • Features: 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：

  1. 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。

  2. 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。

  3. 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。

  4. 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。

  5. "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

  1. ICLR 2026 Official Website
  2. ICLR 2026 Virtual Papers
  3. OpenReview

（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。）

相关推荐
晴天161 小时前
Chrome WebMCP 让网站学会向 AI「自我介绍」
前端·人工智能·chrome
小程故事多_801 小时前
从理论建模到落地运行，深度拆解DDD与本体论的共生与转化逻辑
人工智能·本体论
玩转单片机与嵌入式1 小时前
深入浅出TinyML 24：剪枝、聚类、蒸馏和算子替换是否真正适合MCU？
人工智能·stm32·机器学习·tinyml
byte轻骑兵1 小时前
【BlueZ 】Adapter 模块：蓝牙适配器（主设备）的初始化与核心逻辑
linux·人工智能·bluez·电脑蓝牙·嵌入式蓝牙
Dawson Zhu1 小时前
Kafka 在 AI Agent 系统中的应用：任务队列与事件总线的角色辨析及工程实践
人工智能·语言模型·架构·aigc·agi
黎阳之光1 小时前
AI黑光相机｜赋能低空全域感知，筑牢低空经济全天候视觉防线
人工智能·物联网·算法·安全·数字孪生
wangchunyu1141 小时前
火山引擎 HiAgent介绍和安装说明
人工智能·火山引擎
手写码匠1 小时前
华为云Flexus+DeepSeek征文｜DeepSeek R1 推理优化实战：让复杂任务的回答又快又稳
人工智能·深度学习·算法·aigc
不爱运动的跑者1 小时前
架构可视化不是“画图“：让 AI 产出可追溯架构图的四层工程
人工智能·架构
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）02ZCode 周末送额度活动开启：3 亿 Token 免费领取03GitHub 镜像站点04蜘蛛磁力 搜索引擎大全，如何使用蜘蛛磁力查找磁力链接05历年考研数学一、数学二、数学三真题试卷及答案PDF062026 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？07微信历史版本含下载地址（ Windows PC | 安卓 | MAC ）及设置微信不更新08AI 编程 IDE 全景解析 2026：Agent 全面接管开发链路09Agnes AI 免费 API 接入指南：文本、生图、生视频，一套接口全免费10VSCode+CodeX扩展，无法启动