关注并星标 从此不迷路 计算机视觉研究院 公众号IDComputerVisionGzq 计算机视觉研究院专栏 作者:EdisonG 这个是计算机视觉研究院新推出的模块,后期我们会陆续为大家带来最新文章及技术的代码实现分享! 《TowardsLayerwiseImageVectorization》(CVPR2022) GitHub:github。commaxuLIVE Installation Wesuggestuserstousethecondaforcreatingnewpythonenvironment。 Requirement:5。0GCC6。0;nvcc10。0。gitclonegitgithub。com:maxuLIVE。gitcdLIVEcondacreatenlivepython3。7condaactivatelivecondainstallypytorchtorchvisioncpytorchcondainstallynumpyscikitimagecondainstallycanacondacmakecondainstallyccondaforgeffmpegpipinstallsvgwritesvgpathtoolscssutilsnumbatorchtoolsscikitfmmeasydictvisdompipinstallopencvpython4。5。4。60pleaseinstallthisversiontoavoidsegmentationfault。cdDiffVGgitsubmoduleupdateinitrecursivepythonsetup。pyinstallcd。。 RunExperimentscondaactivatelivecdLIVEPleasemodifytheparamtersaccordingly。pythonmain。pyconfigconfig。yamlexperimentexperimentsettingsignaturegivenfoldernametargetinputimagelogdirlogdirHereisansimpleexample:pythonmain。pyconfigconfigbase。yamlexperimentexperiment5x1signaturesmiletargetfiguressmile。pnglogdirlog 《MultimodalTokenFusionforVisionTransformers》(CVPR2022) GitHub:github。comyikaiwTokenFusion 《PointAugmenting:CrossModalAugmentationfor3DObjectDetection》(CVPR2022) GitHub:github。comVISIONSJTUPointAugmenting 《Fantasticquestionsandwheretofindthem:FairytaleQAAnauthenticdatasetfornarrativecomprehension。》(ACL2022) GitHub:github。comucisoeFairytaleQAData 《LUNAR:UnifyingLocalOutlierDetectionMethodsviaGraphNeuralNetworks》(AAAI2022) GitHub:github。comagoodgeLUNAR Firstly,extractdata。zip ToreplicatetheresultsontheHRSSdatasetwithneighbourcountk100andMixednegativesamplingscheme Extractsavedmodels。zip Run:python3main。pydatasetHRSSsamplesMIXEDk100 Totrainanewmodel:python3main。pydatasetHRSSsamplesMIXEDk100trainnewmodel 《PseudoLabelTransferfromFrameLeveltoNoteLevelinaTeacherStudentFrameworkforSingingTranscriptionfromPolyphonicMusic》(ICASSP2022) GitHub:github。comkeumsicassp2022vocaltranscription 《RobustDisentangledVariationalSpeechRepresentationLearningforZeroshotVoiceConversion》(ICASSP2022) GitHub:github。comjlian2RobustVoiceStyleTransfer Demo:https:jlian2。github。ioRobustVoiceStyleTransfer 《HandoverSim:ASimulationFrameworkandBenchmarkforHumantoRobotObjectHandovers》(ICRA2022) GitHub:github。comNVlabshandoversim2022060316:13:46:Runningevaluationforresults20220228085734yangicra2021s0test2022060316:13:47:Evaluationresults:successratemeanaccumtime(s)failure()()execplantotalhandcontactobjectdroptimeout::::::::::::::64。58(93144)4。8640。0364。90017。36(25144)11。81(17144)6。25(9144)2022060316:13:47:Printingsceneids2022060316:13:47:Success(93scenes):01234567891012131516171819212223252627283033343536373842434649505354565960626364666869707172778183858789919293949596981031061071081091101111121131141151161171201211231251261271281301311321331371381391411432022060316:13:47:Failurehandcontact(25scenes):11142029394041444547515557586567748082881021051181241362022060316:13:47:Failureobjectdrop(17scenes):243132526178798486971011041191221341401422022060316:13:47:Failuretimeout(9scenes):4873757690991001291352022060316:13:47:Evaluationcomplete。 《CDLM:CrossDocumentLanguageModeling》(EMNLP2021) GitHub:github。comavicluCDLM YoucaneitherpretrainbyyourselforusethepretrainedCDLMmodelweightsandtokenizerfiles,whichareavailableonHuggingFace。 Then,use:fromtransformersimportAutoTokenizer,AutoModelloadmodelandtokenizertokenizerAutoTokenizer。frompretrained(biunlpcdlm)modelAutoModel。frompretrained(biunlpcdlm) 《ContinualLearningforTaskOrientedDialogueSystems》(EMNLP2021) GitHub:github。comandreamad8ToDCL 《TorsionalDiffusionforMolecularConformerGeneration》(2022) GitHub:github。comgcorsotorsionaldiffusion 《MMChat:MultiModalChatDatasetonSocialMedia》(2022) GitHub:github。comsilverriverMMChat 《CanCNNsBeMoreRobustThanTransformers?》(2022) GitHub:github。comUCSCVLAARobustCNN 《RevealingSingleFrameBiasforVideoandLanguageLearning》(2022) GitHub:github。comjayleicnsingularity 《ProgressiveDistillationforFastSamplingofDiffusionModels》(2022) GitHub:github。comHramchenkodiffusiondistiller 《NeuralBasisModelsforInterpretability》(2022) GitHub:github。comfacebookresearchnbmspam 《ScalableInterpretabilityviaPolynomials》(2022) GitHub:github。comfacebookresearchnbmspam 《InfiniteRecommendationNetworks:ADataCentricApproach》(2022) GitHub:github。comnoveensinfiniteaecf 《TheGatedTabTransformer。Anenhanceddeeplearningarchitecturefortabularmodeling》(2022) GitHub:github。comradichoGatedTabTransformer Usage:importtorchimporttorch。nnasnnfromgatedtabtransformerimportGatedTabTransformer modelGatedTabTransformer(categories(10,5,6,5,8),tuplecontainingthenumberofuniquevalueswithineachcategorynumcontinuous10,numberofcontinuousvaluestransformerdim32,dimension,papersetat32dimout1,binaryprediction,butcouldbeanythingtransformerdepth6,depth,paperrecommended6transformerheads8,heads,paperrecommends8attndropout0。1,postattentiondropoutffdropout0。1,feedforwarddropoutmlpactnn。LeakyReLU(0),activationforfinalmlp,defaultstorelu,butcouldbeanythingelse(selu,etc。)mlpdepth4,mlphiddenlayersdepthmlpdimension32,dimensionofmlplayersgmlpenabledTruegmlporstandardmlp) xcategtorch。randint(0,5,(1,5))categoryvalues,from0maxnumberofcategories,intheorderaspassedintotheconstructorabovexconttorch。randn(1,10)assumecontinuousvaluesarealreadynormalizedinpidually predmodel(xcateg,xcont)print(pred) 《DistractYourAttention:MultiheadCrossAttentionNetworkforFacialExpressionRecognition》(2022) GitHub:github。comyaoingDAN 《TowardsPrincipledDisentanglementforDomainGeneralization》(2021) GitHub:github。comhlzhang109DDG 《SoundStream:AnEndtoEndNeuralAudioCodec》(2021) GitHub:github。comwesbzSoundStream THEEND 转载请联系本公众号获得授权 计算机视觉研究院学习群等你加入! 计算机视觉研究院主要涉及深度学习领域,主要致力于人脸检测、人脸识别,多目标检测、目标跟踪、图像分割等研究方向。研究院接下来会不断分享最新的论文算法新框架,我们这次改革不同点就是,我们要着重研究。之后我们会针对相应领域分享实践过程,让大家真正体会摆脱理论的真实场景,培养爱动手编程爱动脑思考的习惯! 计算机视觉研究院 公众号IDComputerVisionGzq