caffe神经网络经典结构及模型

我就是我 2022-05-30 01:18 233阅读 0赞

转自博客：[noahsnail.com][] | [CSDN][] | [简书][Link 1]

自2012年Alexnet赢得了ImageNet竞赛以来，深度学习（神经网络）得到了飞速发展，产生了许多的神经网络结构，本文主要总结Caffe中使用的神经网络(分类的神经网络)，本文的神经网络作者都使用Caffe训练过，并在Kaggle的Intel癌症预测比赛中进行了测试与使用（top 8%）。

## 1. Alexnet ##

Alexnet，2012年ImageNet竞赛冠军，深度学习的里程碑。

网络结构地址：[https://github.com/BVLC/caffe/tree/master/models/bvlc\_alexnet][https_github.com_BVLC_caffe_tree_master_models_bvlc_alexnet]

预训练模型地址：[http://dl.caffe.berkeleyvision.org/bvlc\_alexnet.caffemodel][http_dl.caffe.berkeleyvision.org_bvlc_alexnet.caffemodel]

## 2. Squeezenet ##

Squeezenet设计目标不是为了提高识别的准确率，而是希望简化网络复杂度。squeezenet的模型结构确实很小，没压缩的情况下才5M左右，而且识别的精度还可以。

网络结构地址：[https://github.com/DeepScale/SqueezeNet][https_github.com_DeepScale_SqueezeNet]

预训练模型地址：[https://github.com/DeepScale/SqueezeNet][https_github.com_DeepScale_SqueezeNet]

## 3. VGG系列 ##

VGG和GoogLenet是2014年imagenet竞赛的双雄，VGG主要分为VGG16和VGG19。其网络结构与预训练模型的地址如下：

VGG16的网络结构：[https://gist.github.com/ksimonyan/211839e770f7b538e2d8\#file-readme-md][https_gist.github.com_ksimonyan_211839e770f7b538e2d8_file-readme-md]

VGG16的预训练模型：[ http://www.robots.ox.ac.uk/~vgg/software/very\_deep/caffe/VGG\_ILSVRC\_16\_layers.caffemodel][http_www.robots.ox.ac.uk_vgg_software_very_deep_caffe_VGG_ILSVRC_16_layers.caffemodel]

VGG19的网络结构：[https://gist.github.com/ksimonyan/3785162f95cd2d5fee77\#file-readme-md][https_gist.github.com_ksimonyan_3785162f95cd2d5fee77_file-readme-md]

VGG19的预训练模型：[http://www.robots.ox.ac.uk/~vgg/software/very\_deep/caffe/VGG\_ILSVRC\_19\_layers.caffemodel][http_www.robots.ox.ac.uk_vgg_software_very_deep_caffe_VGG_ILSVRC_19_layers.caffemodel]

备注：上面的网络结构需要进行细微调整才能在Caffe中直接训练，主要是网络结构中的Type类型。

## 4. Resnet系列 ##

Resnet网络，2015年ImageNet竞赛冠军，网络结构主要分为Resnet-50、Resnet-101、Resnet-152三种，当然也有一些其它的结构，例如Resnet-18，Resnet-14。

Github地址：[https://github.com/KaimingHe/deep-residual-networks][https_github.com_KaimingHe_deep-residual-networks]

Resnet-50、Resnet-101、Resnet-152的网络结构及预训练模型的下载地址：[https://onedrive.live.com/?authkey=%21AAFW2-FVoxeVRck&id=4006CBB8476FF777%2117887&cid=4006CBB8476FF777][https_onedrive.live.com_authkey_21AAFW2-FVoxeVRck_id_4006CBB8476FF777_2117887_cid_4006CBB8476FF777]

## 5. Inception系列 ##

Inception系列是Google发明的一系列神经网络结构。

Inception-v1：

Inception-v1，即大名鼎鼎的GoogLenet，2014年ImageNet竞赛冠军。

网络结构地址：[https://github.com/BVLC/caffe/tree/master/models/bvlc\_googlenet][https_github.com_BVLC_caffe_tree_master_models_bvlc_googlenet]

预训练模型地址：[http://dl.caffe.berkeleyvision.org/bvlc\_googlenet.caffemodel][http_dl.caffe.berkeleyvision.org_bvlc_googlenet.caffemodel]

Inception-v2：

即Inception V1 + Batch Normalization。

网络结构地址：[https://github.com/pertusa/InceptionBN-21K-for-Caffe][https_github.com_pertusa_InceptionBN-21K-for-Caffe]

预训练模型地址：[http://www.dlsi.ua.es/~pertusa/deep/Inception21k.caffemodel][http_www.dlsi.ua.es_pertusa_deep_Inception21k.caffemodel]

Inception-v3：

网络结构地址：[https://pan.baidu.com/s/1boC0HEf\#list/path=%2F][https_pan.baidu.com_s_1boC0HEf_list_path_2F]

预训练模型地址：[https://pan.baidu.com/s/1boC0HEf\#list/path=%2F][https_pan.baidu.com_s_1boC0HEf_list_path_2F]

Inception-v4：

网络结构地址：[https://pan.baidu.com/s/1c6D150\#list/path=%2F][https_pan.baidu.com_s_1c6D150_list_path_2F]

预训练模型地址：[https://pan.baidu.com/s/1c6D150\#list/path=%2F][https_pan.baidu.com_s_1c6D150_list_path_2F]

Inception-resnet-v2：

网络结构地址：[https://pan.baidu.com/s/1jHPJCX4\#list/path=%2F][https_pan.baidu.com_s_1jHPJCX4_list_path_2F]

预训练模型地址：[https://pan.baidu.com/s/1jHPJCX4\#list/path=%2F][https_pan.baidu.com_s_1jHPJCX4_list_path_2F]

## 6. Densenet系列 ##

网络结构及预训练模型地址：  
[https://github.com/liuzhuang13/DenseNetCaffe][https_github.com_liuzhuang13_DenseNetCaffe]  
[https://github.com/shicai/DenseNet-Caffe][https_github.com_shicai_DenseNet-Caffe]

## Caffe Model Zoo ##

我自己整理的Caffe Model，SENet没有全部添加到仓库，DenseNet没添加到仓库。  
[https://github.com/SnailTyan/caffe-model-zoo][https_github.com_SnailTyan_caffe-model-zoo]

[noahsnail.com]: http://noahsnail.com/
[CSDN]: http://blog.csdn.net/quincuntial
[Link 1]: http://www.jianshu.com/users/7731e83f3a4e/latest_articles
[https_github.com_BVLC_caffe_tree_master_models_bvlc_alexnet]: https://github.com/BVLC/caffe/tree/master/models/bvlc_alexnet
[http_dl.caffe.berkeleyvision.org_bvlc_alexnet.caffemodel]: http://dl.caffe.berkeleyvision.org/bvlc_alexnet.caffemodel
[https_github.com_DeepScale_SqueezeNet]: https://github.com/DeepScale/SqueezeNet
[https_gist.github.com_ksimonyan_211839e770f7b538e2d8_file-readme-md]: https://gist.github.com/ksimonyan/211839e770f7b538e2d8#file-readme-md
[http_www.robots.ox.ac.uk_vgg_software_very_deep_caffe_VGG_ILSVRC_16_layers.caffemodel]: http://www.robots.ox.ac.uk/~vgg/software/very_deep/caffe/VGG_ILSVRC_16_layers.caffemodel
[https_gist.github.com_ksimonyan_3785162f95cd2d5fee77_file-readme-md]: https://gist.github.com/ksimonyan/3785162f95cd2d5fee77#file-readme-md
[http_www.robots.ox.ac.uk_vgg_software_very_deep_caffe_VGG_ILSVRC_19_layers.caffemodel]: http://www.robots.ox.ac.uk/~vgg/software/very_deep/caffe/VGG_ILSVRC_19_layers.caffemodel
[https_github.com_KaimingHe_deep-residual-networks]: https://github.com/KaimingHe/deep-residual-networks
[https_onedrive.live.com_authkey_21AAFW2-FVoxeVRck_id_4006CBB8476FF777_2117887_cid_4006CBB8476FF777]: https://onedrive.live.com/?authkey=%21AAFW2-FVoxeVRck&id=4006CBB8476FF777%2117887&cid=4006CBB8476FF777
[https_github.com_BVLC_caffe_tree_master_models_bvlc_googlenet]: https://github.com/BVLC/caffe/tree/master/models/bvlc_googlenet
[http_dl.caffe.berkeleyvision.org_bvlc_googlenet.caffemodel]: http://dl.caffe.berkeleyvision.org/bvlc_googlenet.caffemodel
[https_github.com_pertusa_InceptionBN-21K-for-Caffe]: https://github.com/pertusa/InceptionBN-21K-for-Caffe
[http_www.dlsi.ua.es_pertusa_deep_Inception21k.caffemodel]: http://www.dlsi.ua.es/~pertusa/deep/Inception21k.caffemodel
[https_pan.baidu.com_s_1boC0HEf_list_path_2F]: https://pan.baidu.com/s/1boC0HEf#list/path=%2F
[https_pan.baidu.com_s_1c6D150_list_path_2F]: https://pan.baidu.com/s/1c6D150#list/path=%2F
[https_pan.baidu.com_s_1jHPJCX4_list_path_2F]: https://pan.baidu.com/s/1jHPJCX4#list/path=%2F
[https_github.com_liuzhuang13_DenseNetCaffe]: https://github.com/liuzhuang13/DenseNetCaffe
[https_github.com_shicai_DenseNet-Caffe]: https://github.com/shicai/DenseNet-Caffe
[https_github.com_SnailTyan_caffe-model-zoo]: https://github.com/SnailTyan/caffe-model-zoo