{"url":"/method/shufflenet-v2-block","slug":"shufflenet-v2-block","name":"ShuffleNet V2 Block","full_name":"ShuffleNet V2 Block","full_name_withheld":false,"description_markdown":"**ShuffleNet V2 Block** is an image model block used in the [ShuffleNet V2](https://paperswithcode.com/method/shufflenet-v2) architecture, where speed is the metric optimized for (instead of indirect ones like FLOPs). It utilizes a simple operator called channel split. At the beginning of each unit, the input of $c$ feature channels are split into two branches with $c - c'$ and $c'$ channels, respectively. Following **G3**, one branch remains as identity. The other branch consists of three convolutions with the same input and output channels to satisfy **G1**. The two $1\\times1$ convolutions are no longer group-wise, unlike the original [ShuffleNet](https://paperswithcode.com/method/shufflenet). This is partially to follow **G2**, and partially because the split operation already produces two groups. After [convolution](https://paperswithcode.com/method/convolution), the two branches are concatenated. So, the number of channels keeps the same (G1). The same “[channel shuffle](https://paperswithcode.com/method/channel-shuffle)” operation as in ShuffleNet is then used to enable information communication between the two branches.\r\n\r\nThe motivation behind channel split is that alternative architectures, where pointwise group convolutions and bottleneck structures are used, lead to increased memory access cost. Additionally more network fragmentation with group convolutions reduces parallelism (less friendly for GPU), and the element-wise addition operation, while they have low FLOPs, have high memory access cost. Channel split is an alternative where we can maintain a large number of equally wide channels (equally wide minimizes memory access cost) without having dense convolutions or too many groups.","description_state":"present","introduced_year":null,"introduced_by":{"title":"ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design","paper":"/paper/shufflenet-v2-practical-guidelines-for","first_author":"Ningning Ma","n_authors":4,"url_abs":null,"archive_paper_url":"https://paperswithcode.com/paper/shufflenet-v2-practical-guidelines-for"},"source":{"url":"http://arxiv.org/abs/1807.11164v1","title":"ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design","url_on_a_paper_host":true},"code_snippet_url":"https://github.com/pytorch/vision/blob/227027d5abc8eacb110c93b5b5c2f4ea5dd401d6/torchvision/models/shufflenetv2.py#L36","code_snippet_url_on_a_code_host":true,"categories":[{"area":"Computer Vision","area_id":"computer-vision","collection":"Image Model Blocks","url":"/methods/category/image-model-blocks","pwc_aliases":[]}],"n_papers_tagged":21,"archive_num_papers":21,"papers_newest_first":[{"paper":null,"title":"Rethinking Information Loss in Medical Image Segmentation with Various-sized Targets","date":"2024-03-28","arxiv_id":"2403.19177","n_code_links":0,"syntology":null},{"paper":null,"title":"Fragility, Robustness and Antifragility in Deep Learning","date":"2023-12-15","arxiv_id":"2312.09821","n_code_links":0,"syntology":null},{"paper":null,"title":"A Non-monotonic Smooth Activation Function","date":"2023-10-16","arxiv_id":"2310.10126","n_code_links":0,"syntology":null},{"paper":null,"title":"Real Time Egocentric Segmentation for Video-self Avatar in Mixed Reality","date":"2022-07-04","arxiv_id":"2207.01296","n_code_links":0,"syntology":null},{"paper":"/paper/smu-smooth-activation-function-for-deep","title":"SMU: smooth activation function for deep networks using smoothing maximum technique","date":"2021-11-08","arxiv_id":"2111.04682","n_code_links":6,"syntology":null},{"paper":null,"title":"SAU: Smooth activation function using convolution with approximate identities","date":"2021-09-27","arxiv_id":"2109.13210","n_code_links":0,"syntology":null},{"paper":null,"title":"ErfAct and Pserf: Non-monotonic Smooth Trainable Activation Functions","date":"2021-09-09","arxiv_id":"2109.04386","n_code_links":0,"syntology":null},{"paper":"/paper/rethinking-image-deraining-via-rain-streaks","title":"Rethinking Image Deraining via Rain Streaks and Vapors","date":"2020-08-03","arxiv_id":"2008.00823","n_code_links":1,"syntology":{"ran":3,"of":3,"unverified":0,"pointer_only":3}},{"paper":null,"title":"Egocentric Human Segmentation for Mixed Reality","date":"2020-05-25","arxiv_id":"2005.12074","n_code_links":0,"syntology":null},{"paper":null,"title":"DyNet: Dynamic Convolution for Accelerating Convolutional Neural Networks","date":"2020-04-22","arxiv_id":"2004.10694","n_code_links":0,"syntology":null},{"paper":"/paper/cnn-cass-cnn-for-classification-of-coronary","title":"CNN-CASS: CNN for Classification of Coronary Artery Stenosis Score in MPR Images","date":"2020-01-23","arxiv_id":"2001.08593","n_code_links":1,"syntology":null},{"paper":null,"title":"Depth-wise Decomposition for Accelerating Separable Convolutions in Efficient Convolutional Neural Networks","date":"2019-10-21","arxiv_id":"1910.09455","n_code_links":0,"syntology":null},{"paper":null,"title":"ThunderNet: Towards Real-Time Generic Object Detection on Mobile Devices","date":"2019-10-01","arxiv_id":null,"n_code_links":0,"syntology":null},{"paper":"/paper/mish-a-self-regularized-non-monotonic-neural","title":"Mish: A Self Regularized Non-Monotonic Activation Function","date":"2019-08-23","arxiv_id":"1908.08681","n_code_links":9,"syntology":{"ran":3,"of":12,"unverified":9,"pointer_only":0}},{"paper":"/paper/dicenet-dimension-wise-convolutions-for","title":"DiCENet: Dimension-wise Convolutions for Efficient Networks","date":"2019-06-08","arxiv_id":"1906.03516","n_code_links":2,"syntology":{"ran":0,"of":1,"unverified":1,"pointer_only":0}},{"paper":"/paper/butterfly-transform-an-efficient-fft-based","title":"Butterfly Transform: An Efficient FFT Based Neural Architecture Design","date":"2019-06-05","arxiv_id":"1906.02256","n_code_links":1,"syntology":null},{"paper":"/paper/thundernet-towards-real-time-generic-object","title":"ThunderNet: Towards Real-time Generic Object Detection","date":"2019-03-28","arxiv_id":"1903.11752","n_code_links":3,"syntology":null},{"paper":"/paper/detnas-neural-architecture-search-on-object","title":"DetNAS: Backbone Search for Object Detection","date":"2019-03-26","arxiv_id":"1903.10979","n_code_links":2,"syntology":{"ran":0,"of":8,"unverified":8,"pointer_only":0}},{"paper":null,"title":"ShuffleNASNets: Efficient CNN models through modified Efficient Neural Architecture Search","date":"2018-12-07","arxiv_id":"1812.02975","n_code_links":0,"syntology":null},{"paper":null,"title":"DSNet for Real-Time Driving Scene Semantic Segmentation","date":"2018-12-06","arxiv_id":"1812.07049","n_code_links":0,"syntology":null},{"paper":"/paper/shufflenet-v2-practical-guidelines-for","title":"ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design","date":"2018-07-30","arxiv_id":"1807.11164","n_code_links":35,"syntology":{"ran":1,"of":30,"unverified":29,"pointer_only":0}}],"papers_shown":21,"tasks":[{"task":"/task/object-detection","name":"Object Detection","papers":7},{"task":"/task/object-detection-1","name":"object-detection","papers":6},{"task":"/task/image-classification","name":"Image Classification","papers":5},{"task":"/task/semantic-segmentation","name":"Semantic Segmentation","papers":5},{"task":"/task/architecture-search","name":"Neural Architecture Search","papers":4},{"task":"/task/classification","name":"General Classification","papers":3},{"task":"/task/segmentation","name":"Segmentation","papers":3},{"task":"/task/image-classification","name":"image-classification","papers":3},{"task":"/task/deep-learning","name":"Deep Learning","papers":2},{"task":"/task/mixed-reality","name":"Mixed Reality","papers":2},{"task":"/task/object","name":"Object","papers":2},{"task":"/task/adversarial-attack","name":"Adversarial Attack","papers":1},{"task":"/task/adversarial-robustness","name":"Adversarial Robustness","papers":1},{"task":"/task/autonomous-driving","name":"Autonomous Driving","papers":1},{"task":"/task/decision-making","name":"Decision Making","papers":1},{"task":"/task/image-generation","name":"Image Generation","papers":1},{"task":"/task/image-restoration","name":"Image Restoration","papers":1},{"task":"/task/image-segmentation","name":"Image Segmentation","papers":1},{"task":"/task/medical-image-segmentation","name":"Medical Image Segmentation","papers":1},{"task":"/task/rain-removal","name":"Rain Removal","papers":1}],"tasks_shown":20,"n_tasks":24,"usage_by_year":[{"year":"2018","papers":3},{"year":"2019","papers":7},{"year":"2020","papers":4},{"year":"2021","papers":3},{"year":"2022","papers":1},{"year":"2023","papers":2},{"year":"2024","papers":1}],"row_source":"methods_table","archive":{"source":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","archive_url":"https://paperswithcode.com/method/shufflenet-v2-block"},"syntology_read_at":"2026-09-24T18:15:14+00:00"}