Building reliable object detectors that are robust to domain shifts, such as various changes in context, viewpoint, and object appearances, is critical for real-world applications. In this work, we study the effectiveness of auxiliary self-supervised tasks to improve the out-of-distribution generalization of object detectors. Inspired by the principle of maximum entropy, we introduce a novel self-supervised task, instance-level temporal cycle confusion (CycConf), which operates on the region features of the object detectors. For each object, the task is to find the most different object proposals in the adjacent frame in a video and then cycle back to itself for self-supervision. CycConf encourages the object detector to explore invariant structures across instances under various motions, which leads to improved model robustness in unseen domains at test time. We observe consistent out-of-domain performance improvements when training object detectors in tandem with self-supervised tasks on large-scale video datasets (BDD100K and Waymo open data). The joint training framework also establishes a new state-of-the-art on standard unsupervised domain adaptative detection benchmarks (Cityscapes, Foggy Cityscapes, and Sim10K). The code and models are available at https://github.com/xinw1012/cycle-confusion.
Beyond Background Shift: Rethinking Instance Replay in Continual Semantic Segmentation
College of Intelligence and Computing, Tianjin University（天津大学智能与计算学部）；New Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences（中国科学院自动化研究所模式识别国家重点实验室）
CDFormer: Cross-Domain Few-Shot Object Detection Transformer Against Feature Confusion
College of Information Science and Electronic Engineering, Zhejiang University（信息科学与电子工程学院，浙江大学）；College of Computer Science and Technology, Zhejiang University（计算机科学与技术学院，浙江大学）；Jinhua Institute of Zhejiang University（浙大金华研究院）
CRISP: Contrastive Residual Injection and Semantic Prompting for Continual Video Instance Segmentation
State Key Laboratory of Robotics and Intelligent Systems, Shenyang Institute of Automation, Chinese Academy of Sciences（机器人与智能系统国家重点实验室，沈阳自动化研究所，中国科学院）；University of Chinese Academy of Sciences（中国科学院大学）；Mohamed bin Zayed University of Artificial Intelligence（穆罕默德·本·扎耶德人工智能大学）
Mask is All You Need: Rethinking Mask R-CNN for Dense and Arbitrary-Shaped Scene Text Detection
Institute of Information Engineering, Chinese Academy of Sciences（中国科学院信息工程研究所）；School of Cyber Security, University of Chinese Academy of Sciences（中国科学院大学网络空间安全学院）；Guangzhou University（广州大学）；Mashang Consumer Finance Co., Ltd.（马上消费金融股份有限公司）
Pseudo-Label Quality Decoupling and Correction for Semi-Supervised Instance Segmentation
Key Laboratory of Multimedia Trusted Perception and Efficient Computing, Ministry of Education of China, Xiamen University, China（中国教育部多媒体可信感知与高效计算重点实验室，厦门大学）；Tencent Youtu Lab（腾讯云视频实验室）
Indian Institute of Technology Hyderabad（印度理工学院海得拉巴分校）；Mohamed bin Zayed University of AI（穆罕默德·本·扎耶德人工智能大学）；Australian National University（澳大利亚国立大学）；Linköping University（林雪平大学）
Which and Where to Focus: A Simple yet Accurate Framework for Arbitrary-Shaped Nearby Text Detection in Scene Images
Institute of Information Engineering, Chinese Academy of Sciences（中国科学院信息工程研究所）；School of Cyber Security, University of Chinese Academy of Sciences（中国科学院大学网络空间安全学院）
Instance-level Visual Active Tracking with Occlusion-Aware Planning
South China University of Technology（南方科技大学）；Pazhou Laboratory（Pazhou实验室）；Key Laboratory of Big Data and Intelligent Robot, Ministry of Education（教育部大数据与智能机器人重点实验室）；University of Chinese Academy of Sciences（中国科学院大学）
ConfusionFlow: A model-agnostic visualization for temporal analysis of classifier confusion
Johannes Kepler University Linz（约翰内斯·开普勒大学林茨）；Imperial College London（帝国理工学院）；Salesbeat GmbH（Salesbeat有限责任公司）；IBM Research（IBM研究院）；The University of British Columbia（不列颠哥伦比亚大学）
Counterfactual Explanation and Instance-Generation using Cycle-Consistent Generative Adversarial Networks
COMSATS University Islamabad（伊斯兰堡COMSATS大学）；National Center of Artificial Intelligence（国家人工智能中心）；University of Strasbourg（斯特拉斯堡大学）；CNRS（法国国家科学研究中心）；Ecole de technologie superieure（高等技术学院）
DExTeR: Weakly Semi-Supervised Object Detection with Class and Instance Experts for Medical Imaging
University of Strasbourg, CNRS, INSERM, ICube, UMR7357, France（斯特拉斯堡大学、法国国家科学研究中心、法国国家医学研究院、ICube、UMR7357、法国）；University Hospital of Strasbourg, France（斯特拉斯堡大学医院、法国）
VISA: VLM-Guided Instance Semantic Auditing for 3D Occupancy World Models
University of Maryland College Park（马里兰大学帕克分校）；Nanjing University of Posts and Telecommunications（南京邮电大学）；Stanford University（斯坦福大学）；Motional AD Inc.（Motional AD公司）
CycAs: Self-supervised Cycle Association for Learning Re-identifiable Descriptions
Tsinghua University（清华大学）；Australian National University（澳大利亚国立大学）；MEGVII Technology（旷视科技）
本文提出了一种自监督学习方法的人重新识别（re-ID）的问题，现有的无监督方法通常依赖于伪标签，如那些从视频tracklets或聚类。使用伪标签的潜在缺点是错误可能累积，并且估计伪ID的数量是具有挑战性的。我们引入了一种不同的无监督方法，使我们能够从原始视频中学习行人嵌入，而无需诉诸伪标签。我们的目标是构建一个自我监督的借口任务，匹配的人重新识别目标。受多目标跟踪中的\emph{data association}概念的启发，我们提出了\textbf{Cyc} le \textbf{As}关联（\textbf{CycAs}）任务：在向前和向后执行一对视频帧之间的数据关联之后，行人实例应该与自身相关联。为了实现这一目标，模型必须学习一个有意义的表示，可以很好地描述框架对中实例之间的对应关系。我们将离散关联过程调整为可微分形式，从而使端到端训练变得可行。实验分两个方面进行：首先，在七个基准测试中，我们将我们的方法与现有的无监督re-ID方法进行了比较，证明了CycAs的优越性。然后，为了进一步验证CycAs在现实应用中的实用价值，我们对自收集的视频进行了训练，并在标准测试集上报告了有希望的性能。
