计算机科学 ›› 2023, Vol. 50 ›› Issue (2): 267-274.doi: 10.11896/jsjkx.220900212
商迪, 吕彦锋, 乔红
SHANG Di, LYU Yanfeng, QIAO Hong
摘要： 增量学习是缩小当前人工智能和人类智能间差距的关键技术,指智能体像人类一样从不稳定数据流中顺序学习多个任务,且不发生遗忘。目标检测是计算机视觉领域的核心任务之一,是计算机理解图像的基石。因此,增量目标检测问题具有重要的研究意义和实际意义。尽管增量学习在图像分类中取得了不错的成果,但基于目标检测的增量学习研究还处于初级阶段。这是因为目标检测相比图像分类更加复杂,它需要同时解决分类和边框回归的问题。不少研究者为解决此问题做了很多努力,但大多数工作都只关注如何保留已学习任务的性能,忽略了模型对新任务的快速适应能力,而这正是增量学习的关键要求。基于大脑的记忆机制,人类可以在学习中不断地提取知识以更好更快地学习新任务,不发生遗忘。受此启发,提出了一种融合编解码记忆重放机制的增量元学习方法。该方法对已学习样本的特征向量进行编码存储和解码重放,从而将不稳定数据流近似为动态稳定数据集,缓解了遗忘问题。同时,设计了一个双循环在线元学习策略,模型在内循环分别基于多批次新旧混合数据进行随机梯度更新,最后在外循环进行元学习,从而获得多任务间的共同结构,使模型具有良好的泛化性能,能够快速适应学习中遇到的新任务。在大型的公开数据集PASCAL VOC 和 MS COCO 上设置了3种增量目标检测实验环境来评估所提算法。实验结果表明,所提算法与最先进的方法相比体现出了具有竞争力的性能,证明了其可以帮助模型更好地抵抗遗忘,具有更好的泛化性能。所提算法基于梯度更新,与模型无关,因此其可以与其他检测框架结合,具有强适应性。
中图分类号:
| [1]MICHAEL M,NEAL J C.Catastrophic interference in connectionist networks:The sequential learning problem[M]//Psychology of Learning and Motivation.Academic Press,1989:109-165. [2]RAIA H,DUSHYANT R,ANDREI A R,et al.Embracingchange:Continual learning in deep neural networks [J].Trends in Cognitive Sciences,2020,24(12):1028-1040. [3]KONSTANTIN S,CORDELIA S,KARTEEK A.Incrementallearning of object detectors without catastrophic forgetting[C]// IEEE International Conference on Computer Vision.2017:3420-3429. [4]HAO Y,FU Y W,JIANG Y G,et al.An end-to end architecture for class-incremental object detection with knowledge distillation[C]// IEEE International Conference on Multimedia and Expo.2019:1-6. [5]CHEN L,YU C Y,CHEN L C.A new knowledge distillation for incremental object detection[C]// International Joint Confe-rence on Neural Networks.2019:1-7. [6]MANO A,TYLER L H,CHRISTOPHER K.RODEO:replayfor online object detection[C]// British Machine Vision Confe-rence.2020. [7]ALAN B.Working memory [J].Science,1992,255:556-559. [8]ALAN B.The episodic buffer:a new component of workingmemory? [J].Trends in Cognitive Sciences,2000,4(11):417-423. [9]PATRICIA S,GOLDMAN R.Regional and cellular fractiona-tion of working memory [J].Proceedings of the National Aca-demy of Sciences of the United States of America,1996,93(24):13473-13480. [10]CHARLE H,SARAH M,GORDON D B.Memory for familiar and unfamiliar words:Evidence for a long-term memory contribution to short-term memory span [J].Journal of Memory and Language,1991,30(6):685-701. [11]PHILIP G,VINCENT F C,SAMUEL S,et al.The long and the short of long-term memory－a molecular framework [J].Nature,1986,322(6078):419-422. [12]ROSS G.Fast r-cnn[C]// IEEE International Conference on Computer Vision.2015:1440-1448. [13]MARK E,AE S M,LUC V G,et al.The pascal visual object classes challenge:A retrospective [J].International Journal of Computer Vision,2015,111(1):98-136. [14]TSUN Y L,MICHAEL M,SERGE B,et al.Microsoft COCO:Common Objects in Context[C]// European Conference on Computer Vision.Berlin:Springer,2014:740-755. [15]DAVUD L,MARC A R.Gradient episodic memory for continual learning[C]// Advances in Neural Information Processing Systems.Curran Associates,Inc,2017:6470-6479. [16]SYLVESTRE A R,ALEXANDER K,GEORG S,et al.icarl:Incremental classifier and representation learning[C]//Procee-dings of the IEEE Conference on Computer Vision and Pattern Recognition.2017:5533-5542. [17]DUSHYANT R,FRANCESCO V,ANDREI A R,et al.Conti-nual unsupervised representation learning[C]//Advances in Neural Information Processing Systems.Curran Associates,Inc,2019:7645-7655. [18]ARSLAN C,ALBERT G,PUNEET K D,et al.Using hindsight to anchor past knowledge in continual learning[C]// Procee-dings of the AAAI Conference on Artificial Intelligence.2021:6993-7001. [19]JAMES K,RAZYAN P,NEIL R,et al.Overcoming catastrophicforgetting in neural networks [J].Proceedings of the National Academy of Sciences of the United States of America,2017,114(13):3521-3526. [20]PRAVENDRA S,VINARY K V,PRATIK M,et al.Calibrating cnns for lifelong learning[C]// Advances in Neural Information Processing Systems.Curran Associates,Inc,2020:15579-15590. [21]JAEHONG Y,SAEHOON K,EUNHO Y,et al.Scalable andorder-robust continual learning with additive parameter decomposition[C]// International Conference on Learning Representations.2020. [22]EKIN D C,BARRET Z,DANDELION M,et al.AutoAugment:Learning Augmentation Policies from Data[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2019:113-123. [23]PENG C,ZHAO K,BRIAN C L.Faster ilod:Incremental lear-ning for object detectors based on faster rcnn [J].PatternRe-cognition Letters,2020,140:109-115. [24]JOSEPH K J J,SALMAN K,FAHAD S K,et al.Towards open world object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2021:5830-5840. [25]ZHANG J T,ZHANG T,SHALINI G,et al.Class-incremental learning via deep model consolidation [C]//Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision.2020. [26]JOSEPH K,JATHUSHAN R,SALMAN K,et al.Incremental object detection via meta-learning [J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2021,44(22):9209-9216. [27]CHELSEA F,PIETER A,SERGEY L.Model agnosticmeta-learning for fast adaptation of deep networks[C]//Proceedings of the 34rd International Conference on Machine Learning.2017:1126-1135. [28]ALEX N,JOSHUA A,JOHN S.On first-order meta-learning algorithms [J].arXiv:1803.02999,2018. [29]ANDREI A R,DUSHYANT R,JAKUB S,et al.Meta-learning with latent embedding optimization[C]//International Confe-rence on Learning Representations.2019. [30]ADAM S,SERGEY B,MATTHEW B,et al.Meta-learning with memory-augmented neural networks[C]//Proceedings of The 33rd International Conference on Machine Learning.2016:20-22. [31]MISHRA N,ROHANINEJAD M,CHEN X,et al.Meta-lear-ning with temporal convolutions [J].arXiv:1707.03141,2017. [32]JAKE S,KEVIN S,RICHARD Z.Prototypical networks forfew-shot learning[C]//Advances in Neural Information Processing Systems.Curran Associates,Inc,2017. [33]FLOOD S,YONGXIN Y,LI Z,et al.Learning to compare:Relation network for few-shot learning[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2018. [34]TIAN Y L,WANG Y,DILIP K,et al.Rethinking few-shotimage classification:A good embedding is all you need?[C]// European Conference on Computer Vision.Berlin:Springer,2020:266-282. [35]WANG T C,RAO M A,HISHAM C,et al.Learning rich features at high-speed for single-shot object detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision.2019. [36]NIE J,RAO M A,HISHAM C,et al.Enriched feature guidedrefinement network for object detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision.2019. [37]PANG Y W,WANG T C,Rao M A,et al.Efficient featurized image pyramid network for single shot detector[C]//Procee-dings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2019. [38]FAHAD S K,XU J L,JOOST V D W,et al.Recognizing actions through action-specific person detection [J].IEEE Transactions on Image Processing,2015,24(11):4422-4432. [39]JIALE C,HISHAM C,RAO M A,et al.D2det:Towards high quality object detection and instance segmentation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2020. [40]HISHAM C,SUN G L,FAHAD S K,et al.Object counting and instance segmentation with image-level supervision[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2019. |
| [1] | 刘航, 普园媛, 吕大华, 赵征鹏, 徐丹, 钱文华. 极化自注意力约束颜色溢出的图像自动上色 Polarized Self-attention Constrains Color Overflow in Automatic Coloring of Image 计算机科学, 2023, 50(3): 208-215. https://doi.org/10.11896/jsjkx.220100149 |
| [2] | 张卫良, 陈秀宏. 跨层融合和感受野扩增的SSD目标检测算法 SSD Object Detection Algorithm with Cross-layer Fusion and Receptive Field Amplification 计算机科学, 2023, 50(3): 231-237. https://doi.org/10.11896/jsjkx.211100281 |
| [3] | 陈亮, 王璐, 李生春, 刘昌宏. 基于深度学习的可视化仪表板生成技术研究 Study on Visual Dashboard Generation Technology Based on Deep Learning 计算机科学, 2023, 50(3): 238-245. https://doi.org/10.11896/jsjkx.230100064 |
| [4] | 华杰, 刘学亮, 赵烨. 基于特征融合的小样本目标检测 Few-shot Object Detection Based on Feature Fusion 计算机科学, 2023, 50(2): 209-213. https://doi.org/10.11896/jsjkx.220500153 |
| [5] | 蔡肖, 陈志华, 盛斌. 基于移位窗口金字塔Transformer的遥感图像目标检测 SPT:Swin Pyramid Transformer for Object Detection of Remote Sensing 计算机科学, 2023, 50(1): 105-113. https://doi.org/10.11896/jsjkx.211100208 |
| [6] | 荣欢, 钱敏峰, 马廷淮, 孙圣杰. 基于先验知识图谱的多代理被遮挡目标类别推理模型 Novel Class Reasoning Model Towards Covered Area in Given Image Based on InformedKnowledge Graph Reasoning and Multi-agent Collaboration 计算机科学, 2023, 50(1): 243-252. https://doi.org/10.11896/jsjkx.220700112 |
| [7] | 刘冬梅, 徐洋, 吴泽彬, 刘倩, 宋斌, 韦志辉. 基于边框距离度量的增量目标检测方法 Incremental Object Detection Method Based on Border Distance Measurement 计算机科学, 2022, 49(8): 136-142. https://doi.org/10.11896/jsjkx.220100132 |
| [8] | 王灿, 刘永坚, 解庆, 马艳春. 基于软标签和样本权重优化的Anchor Free目标检测算法 Anchor Free Object Detection Algorithm Based on Soft Label and Sample Weight Optimization 计算机科学, 2022, 49(8): 157-164. https://doi.org/10.11896/jsjkx.210600240 |
| [9] | 齐秀秀, 王佳昊, 李文雄, 周帆. 基于概率元学习的矩阵补全预测融合算法 Fusion Algorithm for Matrix Completion Prediction Based on Probabilistic Meta-learning 计算机科学, 2022, 49(7): 18-24. https://doi.org/10.11896/jsjkx.210600126 |
| [10] | 祝文韬, 兰先超, 罗唤霖, 岳彬, 汪洋. 改进Faster R-CNN的光学遥感飞机目标检测 Remote Sensing Aircraft Target Detection Based on Improved Faster R-CNN 计算机科学, 2022, 49(6A): 378-383. https://doi.org/10.11896/jsjkx.210300121 |
| [11] | 马宾, 付永康, 王春鹏, 李健, 王玉立. 基于GDIoU损失函数的YOLOv4绝缘子高效定位算法 High Performance Insulators Location Scheme Based on YOLOv4 with GDIoU Loss Function 计算机科学, 2022, 49(6A): 412-417. https://doi.org/10.11896/jsjkx.210600089 |
| [12] | 陈永平, 朱建清, 谢懿, 吴含笑, 曾焕强. 基于外接圆半径差损失的实时安全帽检测算法 Real-time Helmet Detection Algorithm Based on Circumcircle Radius Difference Loss 计算机科学, 2022, 49(6A): 424-428. https://doi.org/10.11896/jsjkx.220100252 |
| [13] | 陈佳舟, 赵熠波, 徐阳辉, 马骥, 金灵枫, 秦绪佳. 三维城市场景中的小物体检测 Small Object Detection in 3D Urban Scenes 计算机科学, 2022, 49(6): 238-244. https://doi.org/10.11896/jsjkx.210400174 |
| [14] | 胡伏原, 万新军, 沈鸣飞, 徐江浪, 姚睿, 陶重犇. 深度卷积神经网络图像实例分割方法研究进展 Survey Progress on Image Instance Segmentation Methods of Deep Convolutional Neural Network 计算机科学, 2022, 49(5): 10-24. https://doi.org/10.11896/jsjkx.210200038 |
| [15] | 沈少朋, 马洪江, 张智恒, 周相兵, 朱春满, 温佐承. 多元时序上状态转移模式的三支漂移检测 Three-way Drift Detection for State Transition Pattern on Multivariate Time Series 计算机科学, 2022, 49(4): 144-151. https://doi.org/10.11896/jsjkx.210600045 |
|
||
首页