{"id":13499116,"url":"https://github.com/hujie-frank/SENet","last_synced_at":"2025-03-29T04:30:34.075Z","repository":{"id":37336528,"uuid":"100795326","full_name":"hujie-frank/SENet","owner":"hujie-frank","description":"Squeeze-and-Excitation Networks","archived":false,"fork":false,"pushed_at":"2019-02-25T08:41:20.000Z","size":1381,"stargazers_count":3379,"open_issues_count":15,"forks_count":839,"subscribers_count":83,"default_branch":"master","last_synced_at":"2024-10-16T02:04:24.328Z","etag":null,"topics":["caffe","gpu","senet"],"latest_commit_sha":null,"homepage":"","language":"Cuda","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"apache-2.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/hujie-frank.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null}},"created_at":"2017-08-19T13:14:53.000Z","updated_at":"2024-10-11T09:24:01.000Z","dependencies_parsed_at":"2022-07-12T12:13:56.337Z","dependency_job_id":null,"html_url":"https://github.com/hujie-frank/SENet","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/hujie-frank%2FSENet","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/hujie-frank%2FSENet/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/hujie-frank%2FSENet/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/hujie-frank%2FSENet/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/hujie-frank","download_url":"https://codeload.github.com/hujie-frank/SENet/tar.gz/refs/heads/master","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":222074991,"owners_count":16926641,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["caffe","gpu","senet"],"created_at":"2024-07-31T22:00:29.190Z","updated_at":"2024-10-31T17:31:33.793Z","avatar_url":"https://github.com/hujie-frank.png","language":"Cuda","funding_links":[],"categories":["Papers\u0026Codes","DLA","1 深度学习基础知识"],"sub_categories":["SENet"],"readme":"# Squeeze-and-Excitation Networks \u003csub\u003e([paper](http://openaccess.thecvf.com/content_cvpr_2018/papers/Hu_Squeeze-and-Excitation_Networks_CVPR_2018_paper.pdf))\u003c/sub\u003e\nBy Jie Hu\u003csup\u003e[1]\u003c/sup\u003e, Li Shen\u003csup\u003e[2]\u003c/sup\u003e, Gang Sun\u003csup\u003e[1]\u003c/sup\u003e.\n\n[Momenta](https://momenta.ai/)\u003csup\u003e[1]\u003c/sup\u003e and [University of Oxford](http://www.robots.ox.ac.uk/~vgg/)\u003csup\u003e[2]\u003c/sup\u003e.\n\n## Approach\n\u003cdiv align=\"center\"\u003e\n  \u003cimg src=\"https://github.com/hujie-frank/SENet/blob/master/figures/SE-pipeline.jpg\"\u003e\n\u003c/div\u003e\n\u003cp align=\"center\"\u003e\n  Figure 1: Diagram of a Squeeze-and-Excitation building block.\n\u003c/p\u003e\n\n\u003cdiv align=\"center\"\u003e\n   \u003cimg src=\"https://github.com/hujie-frank/SENet/blob/master/figures/SE-Inception-module.jpg\" width=\"420\"\u003e\n  \u003cimg src=\"https://github.com/hujie-frank/SENet/blob/master/figures/SE-ResNet-module.jpg\"  width=\"420\"\u003e\n\u003c/div\u003e\n\u003cp align=\"center\"\u003e\n  Figure 2: Schema of SE-Inception and SE-ResNet modules. We set r=16 in all our models.\n\u003c/p\u003e\n\n## Implementation\nIn this repository, Squeeze-and-Excitation Networks are implemented by [Caffe](https://github.com/BVLC/caffe).\n\n### Augmentation\n| Method | Settings |\n|:-:|:-:|\n|Random Mirror| True |\n|Random Crop| 8% ~ 100% |\n|Aspect Ratio | 3/4 ~ 4/3 |\n|Random Rotation| -10° ~ 10°|\n|Pixel Jitter| -20 ~ 20 |\n\n### Note:\n* To achieve efficient training and testing, we combine the consecutive operations ***channel-wise scale*** and ***element-wise summation*** into a single layer **\"Axpy\"** in the architectures with skip-connections, resulting in a considerable reduction in memory cost and computational burden.\n\n* In addition, we found that the implementation for ***global average pooling*** on GPU supported by cuDNN and BVLC/caffe is less efficient. In this regard, we re-implement the operation which achieves significant acceleration.\n\n## Trained Models\n\nTable 1. Single crop validation error on ImageNet-1k (center 224x224 crop from resized image with shorter side = 256). The SENet-154 is one of our superior models used in [ILSVRC 2017 Image Classification Challenge](http://image-net.org/challenges/LSVRC/2017/index) where we won the 1st place (Team name: [WMW](http://image-net.org/challenges/LSVRC/2017/results)).\n\n| Model | Top-1 | Top-5 | Size | Caffe Model | Caffe Model\n|:-:|:-:|:-:|:-:|:-:|:-:|\n|SE-BN-Inception| 23.62 | 7.04 | 46 M| [GoogleDrive](https://drive.google.com/file/d/0BwHV3BlNKkWlTWRRbDZYbVB2WWc/view?usp=sharing) | [BaiduYun](https://pan.baidu.com/s/1qYoPdak)\n|SE-ResNet-50   | 22.37 | 6.36 | 107 M | [GoogleDrive](https://drive.google.com/file/d/0BwHV3BlNKkWlS2QwZHFzM3RjNzg/view?usp=sharing) | [BaiduYun](https://pan.baidu.com/s/1gf5wsLl)\n|SE-ResNet-101  | 21.75  | 5.72 | 189 M | [GoogleDrive](https://drive.google.com/file/d/0BwHV3BlNKkWlTEg4YmcwQ0FoZFU/view?usp=sharing) | [BaiduYun](https://pan.baidu.com/s/1c1FvCWg)\n|SE-ResNet-152  | 21.34  | 5.54 | 256 M | [GoogleDrive](https://drive.google.com/file/d/0BwHV3BlNKkWlcFE0Q2NTcWl3WUE/view?usp=sharing) | [BaiduYun](https://pan.baidu.com/s/1dFEnSzR)\n|SE-ResNeXt-50 (32 x 4d) | 20.97 | 5.54 | 105 M | [GoogleDrive](https://drive.google.com/file/d/0BwHV3BlNKkWlQ2Z0Q204V1RITjA/view?usp=sharing) | [BaiduYun](https://pan.baidu.com/s/1dFbEmbv)\n|SE-ResNeXt-101 (32 x 4d) | 19.81 | 4.96 | 187 M | [GoogleDrive](https://drive.google.com/file/d/0BwHV3BlNKkWleklsNzBiZlprblk/view?usp=sharing) | [BaiduYun](https://pan.baidu.com/s/1qY2wjt6)\n|SENet-154 | 18.68 | 4.47 | 440 M | [GoogleDrive](https://drive.google.com/file/d/0BwHV3BlNKkWlbTFZbzFTSXBUTUE/view?usp=sharing) | [BaiduYun](https://pan.baidu.com/s/1o7HdfAE)\n\nHere we obtain better performance than those reported in the paper.\nWe re-train the SENets described in the paper on a single GPU server with 8 NVIDIA Titan X cards, using a mini-batch of 256 and a initial learning rate of 0.1 with more epoches. \nIn contrast, the results reported in the paper were obtained by training the networks with a larger batch size (1024) and learning rate (0.6) across 4 servers. \n\n## Third-party re-implementations\n0. Caffe. SE-mudolues are integrated with a modificated ResNet-50 using a stride 2 in the 3x3 convolution instead of the first 1x1 convolution which obtains better performance: [Repository](https://github.com/shicai/SENet-Caffe).\n0. TensorFlow. SE-modules are integrated with a pre-activation ResNet-50 which follows the setup in [fb.resnet.torch](https://github.com/facebook/fb.resnet.torch): [Repository](https://github.com/ppwwyyxx/tensorpack/tree/master/examples/ResNet).\n0. TensorFlow. Simple Tensorflow implementation of SENets using Cifar10: [Repository](https://github.com/taki0112/SENet-Tensorflow).\n0. MatConvNet. All the released SENets are imported into [MatConvNet](https://github.com/vlfeat/matconvnet): [Repository](https://github.com/albanie/mcnSENets).\n0. MXNet. SE-modules are integrated with the ResNeXt and more architectures are coming soon: [Repository](https://github.com/bruinxiong/SENet.mxnet).\n0. PyTorch. Implementation of SENets by PyTorch: [Repository](https://github.com/moskomule/senet.pytorch).\n0. Chainer. Implementation of SENets by Chainer: [Repository](https://github.com/nutszebra/SENets).\n## Citation\n\nIf you use Squeeze-and-Excitation Networks in your research, please cite the paper:\n    \n    @inproceedings{hu2018senet,\n      title={Squeeze-and-Excitation Networks},\n      author={Jie Hu and Li Shen and Gang Sun},\n      journal={IEEE Conference on Computer Vision and Pattern Recognition},\n      year={2018}\n    }\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fhujie-frank%2FSENet","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fhujie-frank%2FSENet","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fhujie-frank%2FSENet/lists"}