{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/network-decoupling-from-regular-to-depthwise","title":"Network Decoupling: From Regular to Depthwise Separable Convolutions","arxiv_id":"1808.05517","date":"2018-08-16","proceeding":null,"authors":["Jianbo Guo","Yuxi Li","Weiyao Lin","Yurong Chen","Jianguo Li"],"abstract":"Depthwise separable convolution has shown great efficiency in network design,\nbut requires time-consuming training procedure with full training-set\navailable. This paper first analyzes the mathematical relationship between\nregular convolutions and depthwise separable convolutions, and proves that the\nformer one could be approximated with the latter one in closed form. We show\ndepthwise separable convolutions are principal components of regular\nconvolutions. And then we propose network decoupling (ND), a training-free\nmethod to accelerate convolutional neural networks (CNNs) by transferring\npre-trained CNN models into the MobileNet-like depthwise separable convolution\nstructure, with a promising speedup yet negligible accuracy loss. We further\nverify through experiments that the proposed method is orthogonal to other\ntraining-free methods like channel decomposition, spatial decomposition, etc.\nCombining the proposed method with them will bring even larger CNN speedup. For\ninstance, ND itself achieves about 2X speedup for the widely used VGG16, and\ncombined with other methods, it reaches 3.7X speedup with graceful accuracy\ndegradation. We demonstrate that ND is widely applicable to classification\nnetworks like ResNet, and object detection network like SSD300.","url_abs":"http://arxiv.org/abs/1808.05517v1","url_pdf":"http://arxiv.org/pdf/1808.05517v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"network-decoupling-from-regular-to-depthwise","repo_url":"https://github.com/JianboGuo/network-decoupling","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"caffe2","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"object-detection","task_name":"Object Detection"},{"task_slug":"object-detection-1","task_name":"object-detection"}],"methods":[{"method_slug":"1x1-convolution","method_name":"1x1 Convolution"},{"method_slug":"average-pooling","method_name":"Average Pooling"},{"method_slug":"batch-normalization","method_name":"Batch Normalization"},{"method_slug":"bottleneck-residual-block","method_name":"Bottleneck Residual Block"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"global-average-pooling","method_name":"Global Average Pooling"},{"method_slug":"kaiming-initialization","method_name":"Kaiming Initialization"},{"method_slug":"max-pooling","method_name":"Max Pooling"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"residual-block","method_name":"Residual Block"},{"method_slug":"residual-connection","method_name":"Residual Connection"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1808.05517","atlas_url":"https://app.syntology.ai/?focus=1808.05517","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}