{"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/merging-and-evolution-improving-convolutional","title":"Merging and Evolution: Improving Convolutional Neural Networks for Mobile Applications","arxiv_id":"1803.09127","date":"2018-03-24","proceeding":null,"authors":["Zheng Qin","Zhaoning Zhang","Shiqing Zhang","Hao Yu","Yuxing Peng"],"abstract":"Compact neural networks are inclined to exploit \"sparsely-connected\"\nconvolutions such as depthwise convolution and group convolution for employment\nin mobile applications. Compared with standard \"fully-connected\" convolutions,\nthese convolutions are more computationally economical. However,\n\"sparsely-connected\" convolutions block the inter-group information exchange,\nwhich induces severe performance degradation. To address this issue, we present\ntwo novel operations named merging and evolution to leverage the inter-group\ninformation. Our key idea is encoding the inter-group information with a narrow\nfeature map, then combining the generated features with the original network\nfor better representation. Taking advantage of the proposed operations, we then\nintroduce the Merging-and-Evolution (ME) module, an architectural unit\nspecifically designed for compact networks. Finally, we propose a family of\ncompact neural networks called MENet based on ME modules. Extensive experiments\non ILSVRC 2012 dataset and PASCAL VOC 2007 dataset demonstrate that MENet\nconsistently outperforms other state-of-the-art compact networks under\ndifferent computational budgets. For instance, under the computational budget\nof 140 MFLOPs, MENet surpasses ShuffleNet by 1% and MobileNet by 1.95% on\nILSVRC 2012 top-1 accuracy, while by 2.3% and 4.1% on PASCAL VOC 2007 mAP,\nrespectively.","url_abs":"http://arxiv.org/abs/1803.09127v1","url_pdf":"http://arxiv.org/pdf/1803.09127v1.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":"merging-and-evolution-improving-convolutional","repo_url":"https://github.com/clavichord93/MENet","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"merging-and-evolution-improving-convolutional","repo_url":"https://github.com/osmr/imgclsmob","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"mxnet","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[],"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":"channel-shuffle","method_name":"Channel Shuffle"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"depthwise-convolution","method_name":"Depthwise Convolution"},{"method_slug":"depthwise-separable-convolution","method_name":"Depthwise Separable Convolution"},{"method_slug":"global-average-pooling","method_name":"Global Average Pooling"},{"method_slug":"max-pooling","method_name":"Max Pooling"},{"method_slug":"mobilenetv1","method_name":"MobileNetV1"},{"method_slug":"pointwise-convolution","method_name":"Pointwise Convolution"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"shufflenet","method_name":"ShuffleNet"},{"method_slug":"shufflenet-block","method_name":"ShuffleNet Block"},{"method_slug":"softmax","method_name":"Softmax"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}