Visual features with high potential for generalization are critical for computer vision applications. In addition to the computational overhead associated with layer-by-layer feature stacking to produce multi-scale feature maps, existing approaches also incur high computational costs. To address this issue, we present a compact and efficient scale-in-scale convolution operator called SIS by incorporating an efficient progressive multi-scale architecture into a standard convolu-tion operator. More precisely, the suggested operator uses the channel transform-divide-and-conquer technique to optim-ize conventional channel-wise computing, thereby lowering the computational cost while simultaneously expanding the re-ceptive fields within a single convolution layer. Moreover, the proposed SIS operator incorporates weight-sharing with split-and-interact and recur-and-fuse mechanisms for enhanced variant design. The suggested SIS series is easily plug-gable into any promising convolutional backbone, such as the well-known ResNet and Res2Net. Furthermore, we incor-porated the proposed SIS operator series into 29-layer, 50-layer, and 101-layer ResNet as well as Res2Net variants and evaluated these modified models on the widely used CIFAR, PASCAL VOC, and COCO2017 benchmark datasets, where they consistently outperformed state-of-the-art models on a variety of major vision tasks, including image classification, key point estimation, semantic segmentation, and object detection.

Man Zhou;Xueyang Fu;Aiping Liu

School of Information Science and Tecnology,University of Science and Technology of China,Hefei 230027 China

中国科学技术大学学报

[1]Man Zhou;Xueyang Fu;Aiping Liu-.SIS: A new multi-scale convolutional operator)[J].中国科学技术大学学报,2022(04):56-65

ceptive,gable

SIS,new,multi,scale,convolutional,operator,Visual,features,high,potential,generalization,are,critical,computer,vision,applications,In,addition,computational,overhead,associated,layer,stacking,produce,maps,existing,approaches,also,incur,costs,To,address,this,issue,present,compact,efficient,called,incorporating,progressive,architecture,into,standard,precisely,suggested,uses,channel,transform,divide,conquer,technique,optim,ize,conventional,wise,computing,thereby,lowering,while,simultaneously,expanding,fields,within,single,Moreover,proposed,incorporates,weight,sharing,split,interact,recur,fuse,mechanisms,enhanced,design,series,easily,plug,any,promising,backbone,such,well,known,ResNet,Res2Net,Furthermore,porated,variants,evaluated,these,modified,models,widely,used,CIFAR,PASCAL,VOC,COCO2017,benchmark,datasets,where,they,consistently,outperformed,state,art,variety,major,tasks,including,image,classification,key,point,estimation,semantic,segmentation,object,detection

0.635244