{"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/an-energy-and-gpu-computation-efficient","title":"An Energy and GPU-Computation Efficient Backbone Network for Real-Time Object Detection","arxiv_id":"1904.09730","date":"2019-04-22","proceeding":null,"authors":["Youngwan Lee","Joong-won Hwang","Sangrok Lee","Yuseok Bae","Jongyoul Park"],"abstract":"As DenseNet conserves intermediate features with diverse receptive fields by\naggregating them with dense connection, it shows good performance on the object\ndetection task. Although feature reuse enables DenseNet to produce strong\nfeatures with a small number of model parameters and FLOPs, the detector with\nDenseNet backbone shows rather slow speed and low energy efficiency. We find\nthe linearly increasing input channel by dense connection leads to heavy memory\naccess cost, which causes computation overhead and more energy consumption. To\nsolve the inefficiency of DenseNet, we propose an energy and computation\nefficient architecture called VoVNet comprised of One-Shot Aggregation (OSA).\nThe OSA not only adopts the strength of DenseNet that represents diversified\nfeatures with multi receptive fields but also overcomes the inefficiency of\ndense connection by aggregating all features only once in the last feature\nmaps. To validate the effectiveness of VoVNet as a backbone network, we design\nboth lightweight and large-scale VoVNet and apply them to one-stage and\ntwo-stage object detectors. Our VoVNet based detectors outperform DenseNet\nbased ones with 2x faster speed and the energy consumptions are reduced by 1.6x\n- 4.1x. In addition to DenseNet, VoVNet also outperforms widely used ResNet\nbackbone with faster speed and better energy efficiency. In particular, the\nsmall object detection performance has been significantly improved over\nDenseNet and ResNet.","url_abs":"http://arxiv.org/abs/1904.09730v1","url_pdf":"http://arxiv.org/pdf/1904.09730v1.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":"an-energy-and-gpu-computation-efficient","repo_url":"https://github.com/2anchao/VovJpu","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"an-energy-and-gpu-computation-efficient","repo_url":"https://github.com/Dandelight/VoVNet-PaddlePaddle","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"paddle","reach":null},{"paper_slug":"an-energy-and-gpu-computation-efficient","repo_url":"https://github.com/neilctwu/FCOS-pytorch_Simplified","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"an-energy-and-gpu-computation-efficient","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"}},{"paper_slug":"an-energy-and-gpu-computation-efficient","repo_url":"https://github.com/stigma0617/VoVNet-DeepLabV3","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"an-energy-and-gpu-computation-efficient","repo_url":"https://github.com/stigma0617/VoVNet.pytorch","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"an-energy-and-gpu-computation-efficient","repo_url":"https://github.com/stigma0617/maskrcnn-benchmark-vovnet","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}},{"paper_slug":"an-energy-and-gpu-computation-efficient","repo_url":"https://github.com/suvasis/birdnet2cs231n","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"Apache-2.0"}},{"paper_slug":"an-energy-and-gpu-computation-efficient","repo_url":"https://github.com/vov-net/VoVNet-Detectron","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"an-energy-and-gpu-computation-efficient","repo_url":"https://github.com/vov-net/VoVNet-FCOS","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"NOASSERTION"}},{"paper_slug":"an-energy-and-gpu-computation-efficient","repo_url":"https://github.com/youngwanLEE/VoVNet-RefineDet","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":null},{"paper_slug":"an-energy-and-gpu-computation-efficient","repo_url":"https://github.com/youngwanLEE/vovnet-detectron2","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"NOASSERTION"}}],"tasks":[{"task_slug":null,"task_name":"GPU"},{"task_slug":"object-detection","task_name":"Object Detection"},{"task_slug":"real-time-object-detection","task_name":"Real-Time Object Detection"},{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"},{"task_slug":"small-object-detection","task_name":"Small 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":"concatenated-skip-connection","method_name":"Concatenated Skip Connection"},{"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":"mask-r-cnn","method_name":"Mask R-CNN"},{"method_slug":"max-pooling","method_name":"Max Pooling"},{"method_slug":"one-shot-aggregation","method_name":"One-Shot Aggregation"},{"method_slug":"rpn","method_name":"RPN"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"residual-block","method_name":"Residual Block"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"roi-align","method_name":"RoIAlign"},{"method_slug":"speed","method_name":"SPEED"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"step-decay","method_name":"Step Decay"},{"method_slug":"vovnet","method_name":"VoVNet"}],"datasets_introduced":[],"methods_introduced":[{"slug":"one-shot-aggregation","name":"One-Shot Aggregation","full_name":"One-Shot Aggregation"}],"results":[{"leaderboard":"/sota/instance-segmentation-on-coco","task":"Instance Segmentation","dataset":"COCO test-dev","model":"VoVNetV1-57","rank_in_archive_order":66,"of":112,"metrics":{"mask AP":"40.8%"},"uses_additional_data":false},{"leaderboard":"/sota/instance-segmentation-on-coco","task":"Instance Segmentation","dataset":"COCO test-dev","model":"VoVNetV1-39","rank_in_archive_order":75,"of":112,"metrics":{"mask AP":"39.7%"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1904.09730","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}