本文首发极市平台公众号,转载请获得授权并标明出处。 ECCV2022论文分方向整理目前在极市社区持续更新中,已累计更新了54篇,项目地址:https:github。comextremeassistantECCV2022PaperCodeInterpretation 以下是本周更新的ECCV2022论文,包含检测,分割,图像处理,视频理解,神经网络结构设计,无监督学习,迁移学习等方向。 论文合集打包下载地址:https:www。cvmart。netcommunitydetail6592 检测 分割 图像处理 视频处理 图像、视频检索与理解 估计 目标跟踪 文本检测与识别 GAN生成式对抗式 神经网络结构设计 数据处理 模型训练泛化 模型压缩 模型评估 半监督学习自监督学习 多模态跨模态学习 小样本学习 强化学习检测2D目标检测 〔1〕PointtoBoxNetworkforAccurateObjectDetectionviaSinglePointSupervision(通过单点监督实现精确目标检测的点对盒网络) paper:https:arxiv。orgabs2207。06827 code:https:github。comucasvgp2bnet 〔2〕YouShouldLookatAllObjects(您应该查看所有物体) paper:https:arxiv。orgabs2207。07889 code:https:github。comcharlespikachuyslao 〔3〕AdversariallyAwareRobustObjectDetector(对抗性感知鲁棒目标检测器) paper:https:arxiv。orgabs2207。06202 code:https:github。com7eu7d7robustdet 3D目标检测 〔1〕RethinkingIoUbasedOptimizationforSinglestage3DObjectDetection(重新思考基于IoU的单阶段3D对象检测优化) paper:https:arxiv。orgabs2207。09332 人物交互检测 〔1〕TowardsHardPositiveQueryMiningforDETRbasedHumanObjectInteractionDetection(面向基于DETR的人机交互检测的硬性查询挖掘) paper:https:arxiv。orgabs2207。05293 code:https:github。commuchhairhqm 图像异常检测 〔1〕DICE:LeveragingSparsificationforOutofDistributionDetection(DICE:利用稀疏化进行分布外检测) paper:https:arxiv。orgabs2111。09805 code:https:github。comdeeplearningwiscdice分割实例分割 〔1〕BoxsupervisedInstanceSegmentationwithLevelSetEvolution(具有水平集进化的框监督实例分割) paper:https:arxiv。orgabs2207。09055 〔2〕OSFormer:OneStageCamouflagedInstanceSegmentationwithTransformers(OSFormer:使用Transformers进行单阶段伪装实例分割) paper:https:arxiv。orgabs2207。02255 code:https:github。compjlallenosformer 语义分割 〔1〕2DPASS:2DPriorsAssistedSemanticSegmentationonLiDARPointClouds(2DPASS:激光雷达点云上的二维先验辅助语义分割) paper:https:arxiv。orgabs2207。04397 code:https:github。comyanx272dpass视频目标分割 〔1〕LearningQualityawareDynamicMemoryforVideoObjectSegmentation(视频对象分割的学习质量感知动态内存) paper:https:arxiv。orgabs2207。07922 code:https:github。comworkforaiqdmn 图像处理超分辨率 〔1〕DynamicDualTrainableBoundsforUltralowPrecisionSuperResolutionNetworks(超低精度超分辨率网络的动态双可训练边界) paper:https:arxiv。orgabs2203。03844 code:https:github。comzysxmuddtb图像去噪 〔1〕DeepSemanticStatisticsMatching(D2SM)DenoisingNetwork(深度语义统计匹配(D2SM)去噪网络) paper:https:arxiv。orgabs2207。09302图像复原图像增强图像重建 〔1〕SemanticSparseColorizationNetworkforDeepExemplarbasedColorization(用于基于深度示例的着色的语义稀疏着色网络) paper:https:arxiv。orgabs2112。01335 〔2〕GeometryawareSingleimageFullbodyHumanRelighting(几何感知单图像全身人体重新照明) paper:https:arxiv。orgabs2207。04750 〔3〕MultiModalMaskedPreTrainingforMonocularPanoramicDepthCompletion(单目全景深度补全的多模态蒙面预训练) paper:https:arxiv。orgabs2203。09855 〔4〕PanoFormer:PanoramaTransformerforIndoor360DepthEstimation(PanoFormer:用于室内360深度估计的全景变压器) paper:https:arxiv。orgabs2203。09283 〔5〕SESS:SaliencyEnhancingwithScalingandSliding(SESS:通过缩放和滑动增强显着性) paper:https:arxiv。orgabs2207。01769 〔6〕RigNet:RepetitiveImageGuidedNetworkforDepthCompletion(RigNet:用于深度补全的重复图像引导网络) paper:https:arxiv。orgabs2107。13802 图像外推(ImageOutpainting) 〔1〕OutpaintingbyQueries(通过查询进行外包) paper:https:arxiv。orgabs2207。05312 code:https:github。comkaiseemqueryotr 风格迁移(StyleTransfer) 〔1〕CCPL:ContrastiveCoherencePreservingLossforVersatileStyleTransfer(CCPL:通用风格迁移的对比相干性保留损失) paper:https:arxiv。orgabs2207。04808 code:https:github。comJarrentWu1031CCPL视频处理(VideoProcessing) 〔1〕ImprovingthePerceptualQualityof2DAnimationInterpolation(提高二维动画插值的感知质量) paper:https:arxiv。orgabs2111。12792 code:https:github。comshuhongcheneisaianimeinterpolator 〔2〕RealTimeIntermediateFlowEstimationforVideoFrameInterpolation(视频帧插值的实时中间流估计) paper:https:arxiv。orgabs2011。06294 code:https:github。comMegEnginearXiv2020RIFE图像、视频检索与理解动作识别 〔1〕ReAct:TemporalActionDetectionwithRelationalQueries(ReAct:使用关系查询的时间动作检测) paper:https:arxiv。orgabs2207。07097 code:https:github。comsssstereact 〔2〕HuntingGroupClueswithTransformersforSocialGroupActivityRecognition(用Transformers寻找群体线索用于社会群体活动识别) paper:https:arxiv。orgabs2207。05254视频理解 〔1〕GraphVid:ItOnlyTakesaFewNodestoUnderstandaVideo(GraphVid:只需几个节点即可理解视频) paper:https:arxiv。orgabs2207。01375 〔2〕DeepHashDistillationforImageRetrieval(用于图像检索的深度哈希蒸馏) paper:https:arxiv。orgabs2112。08816 code:https:github。comyoungkyunjangdeephashdistillation 视频检索(VideoRetrieval) 〔1〕TS2Net:TokenShiftandSelectionTransformerforTextVideoRetrieval(TS2Net:用于文本视频检索的令牌移位和选择转换器) paper:https:arxiv。orgabs2207。07852 code:https:github。comyuqi657ts2net 〔2〕LightweightAttentionalFeatureFusion:ANewBaselineforTexttoVideoRetrieval(轻量级注意力特征融合:文本到视频检索的新基线) paper:https:arxiv。orgabs2112。01832估计位姿估计 〔1〕CategoryLevel6DObjectPoseandSizeEstimationusingSelfSupervisedDeepPriorDeformationNetworks(使用自监督深度先验变形网络的类别级6D对象姿势和大小估计) paper:https:arxiv。orgabs2207。05444 code:https:github。comjiehonglinselfdpdn 深度估计 〔1〕PhysicalAttackonMonocularDepthEstimationwithOptimalAdversarialPatches(使用最优对抗补丁对单目深度估计进行物理攻击) paper:https:arxiv。orgabs2207。04718目标跟踪 〔1〕TowardsGrandUnificationofObjectTracking(迈向目标跟踪的大统一) paper:https:arxiv。orgabs2207。07078 code:https:github。commasterbiniiauunicorn 文本检测与识别 〔1〕DynamicLowResolutionDistillationforCostEfficientEndtoEndTextSpotting(用于经济高效的端到端文本识别的动态低分辨率蒸馏) paper:https:arxiv。orgabs2207。06694 code:https:github。comhikopensourcedavarlabocrGAN生成式对抗式 〔1〕EliminatingGradientConflictinReferencebasedLineArtColorization(消除基于参考的艺术线条着色中的梯度冲突) paper:https:arxiv。orgabs2207。06095 code:https:github。comkunkun0w0sga 〔2〕WaveGAN:FrequencyawareGANforHighFidelityFewshotImageGeneration(WaveGAN:用于高保真少镜头图像生成的频率感知GAN) paper:https:arxiv。orgabs2207。07288 code:https:github。comkobeshegueccv2022wavegan 〔3〕FakeCLR:ExploringContrastiveLearningforSolvingLatentDiscontinuityinDataEfficientGANs(FakeCLR:探索对比学习以解决数据高效GAN中的潜在不连续性) paper:https:arxiv。orgabs2207。08630 code:https:github。comiceli1007fakeclr 〔4〕UniCR:UniversallyApproximatedCertifiedRobustnessviaRandomizedSmoothing(UniCR:通过随机平滑获得普遍近似的认证鲁棒性) paper:https:arxiv。orgabs2207。02152神经网络结构设计神经网络架构搜索(NAS) 〔1〕ScaleNet:SearchingfortheModeltoScale(ScaleNet:搜索要扩展的模型) paper:https:arxiv。orgabs2207。07267 code:https:github。comluminolxscalenet 〔2〕EnsembleKnowledgeGuidedSubnetworkSearchandFinetuningforFilterPruning(集成知识引导的子网络搜索和过滤器修剪微调) paper:https:arxiv。orgabs2203。02651 code:https:github。comsseung0703ekg 〔3〕EAGAN:EfficientTwostageEvolutionaryArchitectureSearchforGANs(EAGAN:GAN的高效两阶段进化架构搜索) paper:https:arxiv。orgabs2111。15097 code:https:github。commarsggboEAGAN数据处理归一化 〔1〕FinegrainedDataDistributionAlignmentforPostTrainingQuantization(训练后量化的细粒度数据分布对齐) paper:https:arxiv。orgabs2109。04186 code:https:github。comzysxmufdda模型训练泛化噪声标签 〔1〕LearningwithNoisyLabelsbyEfficientTransitionMatrixEstimationtoCombatLabelMiscorrection(通过有效的转移矩阵估计学习噪声标签以对抗标签错误校正) paper:https:arxiv。orgabs2111。14932 模型压缩知识蒸馏 〔1〕KnowledgeCondensationDistillation(知识浓缩蒸馏) paper:https:arxiv。orgabs2207。05409 code:https:github。comdzy3kcd)模型评估 〔1〕HierarchicalLatentStructureforMultiModalVehicleTrajectoryForecasting(多模式车辆轨迹预测的分层潜在结构) paper:https:arxiv。orgabs2207。04624 code:https:github。comd1024choihlstrajforecast半监督学习无监督学习自监督学习 〔1〕FedX:UnsupervisedFederatedLearningwithCrossKnowledgeDistillation(FedX:具有交叉知识蒸馏的无监督联合学习) paper:https:arxiv。orgabs2207。09158 〔2〕SynergisticSelfsupervisedandQuantizationLearning(协同自监督和量化学习) paper:https:arxiv。orgabs2207。05432 code:https:github。commegviiresearchssqleccv2022) 〔3〕ContrastiveDeepSupervision(对比深度监督) paper:https:arxiv。orgabs2207。05306 code:https:github。comarchiplablinfengzhangcontrastivedeepsupervision 〔4〕DenseTeacher:DensePseudoLabelsforSemisupervisedObjectDetection(稠密教师:用于半监督目标检测的稠密伪标签) paper:https:arxiv。orgabs2207。02541 〔5〕ImageCodingforMachineswithOmnipotentFeatureLearning(具有全能特征学习的机器的图像编码) paper:https:arxiv。orgabs2207。01932多模态学习跨模态视觉语言 〔1〕ContrastiveVisionLanguagePretrainingwithLimitedResources(资源有限的对比视觉语言预训练) paper:https:arxiv。orgabs2112。09331 code:https:github。comzerovlzerovl跨模态 〔1〕CrossmodalPrototypeDrivenNetworkforRadiologyReportGeneration(用于放射学报告生成的跨模式原型驱动网络) paper:https:arxiv。orgabs code:https:github。commarkinwangxpronet 小样本学习 〔1〕LearningInstanceandTaskAwareDynamicKernelsforFewShotLearning(用于少数镜头学习的学习实例和任务感知动态内核) paper:https:arxiv。orgabs2112。03494 迁移学习自适应 〔1〕FactorizingKnowledgeinNeuralNetworks(在神经网络中分解知识) paper:https:arxiv。orgabs2207。03337 code:https:github。comadamdadknowledgefactor 〔2〕CycDA:UnsupervisedCycleDomainAdaptationfromImagetoVideo(CycDA:从图像到视频的无监督循环域自适应) paper:https:arxiv。orgabs2203。16244 强化学习 〔1〕TargetabsentHumanAttention(目标缺失人类注意力缺失) paper:https:arxiv。orgabs2207。01166 code:https:github。comneouyghursess