Grouped Pointwise Convolutions Reduce Parameters in Convolutional Neural Networks

dc.contributor.authorSchwarz Schuler, Joao Paulo
dc.contributor.authorRomani, Santiago
dc.contributor.authorAbdel-Nasser, Mohamed
dc.contributor.authorRashwan, Hatem
dc.contributor.authorPuig, Domenec
dc.coverage.issue1cs
dc.coverage.volume28cs
dc.date.accessioned2022-06-30T07:01:57Z
dc.date.available2022-06-30T07:01:57Z
dc.date.issued2022-06-30cs
dc.description.abstractIn DCNNs, the number of parameters in pointwise convolutions rapidly grows due to the multiplication of the number of filters by the number of input channels that come from the previous layer. Our proposal makes pointwise convolutions parameter efficient via grouping filters into parallel branches or groups, where each branch processes a fraction of the input channels. However, by doing so, the learning capability of the DCNN is degraded. To avoid this effect, we suggest interleaving the output of filters from different branches at intermediate layers of consecutive pointwise convolutions. We applied our improvement to the EfficientNet, DenseNet-BC L100, MobileNet and MobileNet V3 Large architectures. We trained these architectures with the CIFAR-10, CIFAR-100, Cropped-PlantDoc and The Oxford-IIIT Pet datasets. When training from scratch, we obtained similar test accuracies to the original EfficientNet and MobileNet V3 Large architectures while saving up to 90% of the parameters and 63% of the flops.en
dc.formattextcs
dc.format.extent23-31cs
dc.format.mimetypeapplication/pdfen
dc.identifier.citationMendel. 2022 vol. 28, č. 2, s. 23-31. ISSN 1803-3814cs
dc.identifier.doi10.13164/mendel.2022.1.023en
dc.identifier.issn2571-3701
dc.identifier.issn1803-3814
dc.identifier.urihttp://hdl.handle.net/11012/208126
dc.language.isoencs
dc.publisherInstitute of Automation and Computer Science, Brno University of Technologycs
dc.relation.ispartofMendelcs
dc.relation.urihttps://mendel-journal.org/index.php/mendel/article/view/169cs
dc.rightsCreative Commons Attribution-NonCommercial-ShareAlike 4.0 International licenseen
dc.rights.accessopenAccessen
dc.rights.urihttp://creativecommons.org/licenses/by-nc-sa/4.0en
dc.subjectEfficientNeten
dc.subjectDeep Learningen
dc.subjectComputer Visionen
dc.subjectCNNen
dc.subjectDCNNen
dc.titleGrouped Pointwise Convolutions Reduce Parameters in Convolutional Neural Networksen
dc.type.driverarticleen
dc.type.statusPeer-revieweden
dc.type.versionpublishedVersionen
eprints.affiliatedInstitution.facultyFakulta strojního inženýrstvícs
Files
Original bundle
Now showing 1 - 1 of 1
Loading...
Thumbnail Image
Name:
169-Article Text-408-2-10-20220629.pdf
Size:
2.68 MB
Format:
Adobe Portable Document Format
Description:
Collections