ThePyTorchimplementationofDB-CNNisreleasedathttps://github.com/zwx8981/DBCNN-PyTorch!
BlindImageQualityAssessmentUsingADeepBilinearConvolutionalNeuralNetworkWeixiaZhang,KedeMa,JiaYan,DexiangDeng,andZhouWanghttps://ieeexplore.ieee.org/document/8576582
IEEETransactionsonCircuitsandSystemsforVideoTechnology(TCSVT),Volume:30,Issue:1,Jan.2020.
Filesunderdistorion_generatorareusedforsynthesizingdistortedimages.Usuage:
distorted_img=distortion_generator(img,dist_type,level,seed)
Whereimgistheoriginalpristineimage,dist_typereferstoaspecifieddistortiontyperangingin1~9.
1,GaussianBlur2,WhiteNoise3,JPEGCompression4,JPEG2000Compression5,ContrastChange6,PinkNoise7,ImageColorQuantizationwithDither8,Over-Exposure9,Under-Exposure
levelisaspecifieddegradationlevelrangein1~5.
seedshouldbefixedtobe1.
Trainingcodesliveindbcnnfolder.Runningtherun_exp.mscripttotrainandtestonaspecifiddatasetacross10randomsplits.
Prerequisite:Matlab(Weuse2017a),MatConvNet(Weuse1.0-beta25),vlfeat(Weuse0.9.2)
Pretraineds-cnnmodelisincludedindbcnn\data\models,youshoulddownloadvgg-16modelfromhttps://www.vlfeat.org/matconvnet/pretrained/andputitindbcnn\data\models.
Youneedtocopythematconvet/matlabfoldertothatofyourmatconvnettomodifythevl_simplenn.mandPDist.mfiles.Relevantlinks:WaterlooExplorationDatabase:https://ece.uwaterloo.ca/~k29ma/exploration/PASCALVOC2012:https://host.robots.ox.ac.uk/pascal/VOC/voc2012/
Citation@article{zhang2020blind,title={BlindImageQualityAssessmentUsingADeepBilinearConvolutionalNeuralNetwork},author={Zhang,WeixiaandMa,KedeandYan,JiaandDeng,DexiangandWang,Zhou},journal={IEEETransactionsonCircuitsandSystemsforVideoTechnology},volume={30},number={1},pages={36--47},year={2020}}
评论