caffe笔记

一时失言乱红尘 2022-09-25 11:30 162阅读 0赞

[Caffe各版本与源码全透析][Caffe]

[ubuntu14.04下CPU的caffe配置][ubuntu14.04_CPU_caffe]

[Caffe+ubuntu14.04+Cuda8.0安装][Caffe_ubuntu14.04_Cuda8.0]

http://www.cnblogs.com/denny402/tag/caffe/

[关于caffe里面image\_data\_layer的使用][caffe_image_data_layer]

[Caffe学习系列(11)：图像数据转换成db（leveldb/lmdb)文件][Caffe_11_db_leveldb_lmdb]

[caffe神经网络框架的辅助工具（将图片转换为leveldb格式）][caffe_leveldb]

[将图像转换成lmdb格式][lmdb]

[Python lmdb][]

[Mnist数据集创建lmdb或leveldb类型的数据][Mnist_lmdb_leveldb]

[Caffe1——Mnist数据集创建lmdb或leveldb类型的数据][Mnist_lmdb_leveldb][  
][lmdb]

[Caffe2——cifar10数据集创建lmdb或leveldb类型的数据][Caffe2_cifar10_lmdb_leveldb]

[Caffe3——ImageNet数据集创建lmdb类型的数据][Caffe3_ImageNet_lmdb]

[caffe 数据库LMDB的读写][caffe _LMDB]

[计算图像均值][Link 1]

[【Caffe代码解析】compute\_image\_mean][Caffe_compute_image_mean]

[【Caffe代码解析】convert\_imageset][Caffe_convert_imageset]

[制作train.txt和val.txt文件][train.txt_val.txt]

[txt和lmdb训练caffenet设置对比][txt_lmdb_caffenet]

[计算图像均值][Link 2]

[Caffe训练过程：test\_iter test\_interval等概念][Caffe_test_iter test_interval]

[关于 epoch、 iteration和batchsize][epoch_ iteration_batchsize]

[caffe层解读系列——BatchNorm][caffe_BatchNorm]

[Mnist手写数字识别测试与分析][Mnist]

[修改lenet网络进行训练][lenet]

[Caffe模型的结构][Caffe 1]

[数据层及参数设置][Link 3]

[数据层及参数][Link 4]

[视觉层及参数][Link 5]

[损失层及其参数设置][Link 6]

[其他常用层及参数][Link 7]

[激活函数][Link 8]

[Blob,Layer和Net以及对应配置文件的编写][Blob_Layer_Net]

[命令行解析][Link 9]

[训练和测试自己的图片][Link 10]

[caffe的总体流程是怎样的？][caffe]

[caffe层解读系列-softmax\_loss][caffe_-softmax_loss]

[caffe层解读系列——hinge\_loss][caffe_hinge_loss]

[人脸识别：PairLoss][PairLoss]

[Windows下用VS2013加载caffemodel做图像分类][Windows_VS2013_caffemodel]

[修改网络并训练模型][Link 11]

[caffe基础][caffe 1]

[如何在Caffe中配置每一个层的结构][Caffe 2]

[caffe学习笔记][caffe 2]

[caffe 参数介绍][caffe 3]

[solver优化方法][solver]

[怎么在Caffe中配置每一个层的结构][Caffe 3]

[【Caffe实践】 多目标输出探究][Caffe_]

[封装caffe-windows-master成动态链接库][caffe-windows-master]

[如何快糙好猛地在Windows下编译CAFFE并使用其matlab和python接口][Windows_CAFFE_matlab_python]

[Caffe使用教程][Caffe 4]

[What is 'lr\_policy' in caffe?][What is _lr_policy_ in caffe]

[使用caffe训练时Loss变为nan的原因][caffe_Loss_nan]

[Net Surgery][]

[Caffe中的各种loss函数适合哪些问题？][Caffe_loss]

[fine\_tuning原理问题][fine_tuning]

[Fine-tuning CaffeNet for Style Recognition on “Flickr Style” Data][Fine-tuning CaffeNet for Style Recognition on _Flickr Style_ Data]

[假如有两个model，如何把其中一个model的某一层的权重赋给另一个model的对应层。][model_model_model]

[怎样用自己的数据集对caffe训练好的model进行fineture？][caffe_model_fineture]

[Caffe 是否支持多GPU并行训练？][Caffe _GPU]

[caffe的proto里面的std是什么意思？怎么调？][caffe_proto_std]

[global average pooling 与average pooling的区别是什么？][global average pooling _average pooling]

[各种Pooling方法适用哪些情况？][Pooling]

[如何通过loss曲线判断网络训练的情况][loss]

[无监督逐层预训练目前是否还广泛应用于CNN的训练？][CNN]

[无监督学习是否被监督学习取代了？][Link 12]

[应该怎么理解End to End 训练][End to End]

[caffe中image\_mean的作用是什么？][caffe_image_mean]

[出现过拟合的根本原因是什么][Link 13]

[目前state of art 的人脸检测算法都是cnn based，如何应用于手机等嵌入式设备中？][state of art _cnn based]

[caffe在训练过程中可以不使用validation吗？不使用有什么坏处（或者说使用validation有什么好处）？][caffe_validation_validation]

http://blog.csdn.net/u012235274/article/category/6197462

[深层学习为何要“Deep”（上）][Deep]

## caffe源码 ##

[如何解读Caffe源码][Caffe 5]

[caffe源码解析 — caffe.proto][caffe_ _ caffe.proto]

[caffe数据格式（Google Protocol Buffers）][caffe_Google Protocol Buffers]

[Caffe常用层参数介绍][Caffe 6]

[Caffe源码解析2：SycedMem][Caffe_2_SycedMem]

[caffe源码解析 — SyncedMemory.cpp][caffe_ _ SyncedMemory.cpp]

http://blog.csdn.net/seven\_first/article/category/5721883

[在可视化环境下进行Caffe网络设计][Caffe 7]

[caffe基础(8):draw\_net.py绘制网络结构][caffe_8_draw_net.py]

[caffe示例实现之10LeNet的python接口][caffe_10LeNet_python]

[caffe的python接口封装原理与解析][caffe_python]

[pycaffe学习][pycaffe]

[pycaffe][pycaffe 1]

[caffe python layer][]

[caffe中使用python layer][caffe_python layer]

添加[caffe python][] layer

在python中[caffe示例实现之11逻辑回归及其深化][caffe_11]

[利用caffe的python接口进行训练时遇到的问题][caffe_python 1]

[用 caffe做图像分割实验时，loss值很诡异][caffe_loss]

[caffe 练习4 ----利用python批量抽取caffe计算得到的特征------by 香蕉麦乐迪][caffe _4 ----_python_caffe_------by]

[caffe 练习3 用caffe提供的C++函数批量抽取图像特征------by 香蕉麦乐迪][caffe _3 _caffe_C_------by]

[训练一个全卷积网络，怎么做？][Link 14]

[caffe调试 ubuntu1404+eclipse][caffe_ ubuntu1404_eclipse]

[（Caffe，LeNet）网络训练流程（二）][Caffe_LeNet]

[（Caffe，LeNet）反向传播（六）][Caffe_LeNet 1]

[利用Caffe做回归(regression)][Caffe_regression]

训练可视化

[caffe 画损失曲线和准确率曲线][caffe 4]

[cnn 调参经验][cnn]

可视化

[神经网络：caffe特征可视化的代码样例][caffe 5]

[caffe源码修改：抽取任意一张图片的特征][caffe 6]

[在caffe 中添加Crowd counting 数据层][caffe _Crowd counting]

[在 caffe 中添加 FaceNet 中 Triplet Loss Layer][caffe _ FaceNet _ Triplet Loss Layer]

[caffe 里面GPU 选择设置， 通过配置文件 读取GPU 的编号。][caffe _GPU _ _ _GPU]

[caffe adaboost][]

[Linux下Caffe如何调试][Linux_Caffe]

[\*\_train\_test.prototxt,\*\_deploy.prototxt,\*\_slover.prototxt文件编写时注意事项][train_test.prototxt_deploy.prototxt_slover.prototxt]

关于propagate\_down

比如有4个全连接层A->B->C->D  
    a. 你希望C层的参数不会改变，C前面的AB层的参数也不会改变，这种情况也就是D层的梯度不往前反向传播到D层的输入blob（也就是C层的输出blob 没有得到梯度），你可以通过设置D层的propagate\_down为false来做到。  
         propagate\_down的数量与输入blob的数量相同，假如你某个层有2个输入blob，那么你应该在该layer的Param里面写上两行：  
         propagate\_down : 0    \# 第1个输入blob不会得到反向传播的梯度  
         propagate\_down : 0    \# 第2个输入blob不会得到反向传播的梯度  
         这样的话，你这个layer的梯度就不会反向传播啦，前面的所有layer的参数也就不会改变了  
    b. 你希望C层的参数不会改变，但是C前面的AB层的参数会改变，这种情况，只是固定了C层的参数，C层得到的梯度依然会反向传播给前面的B层。只需要将对应的参数blob的学习率调整为0：  
你在layer里面加上param \{ lr\_mult: 0 \}就可以了，比如全连接层里面：  
layer \{  
    type: "InnerProduct"  
    param \{ \# 对应第1个参数blob的配置，也就是全连接层的参数矩阵的配置  
         lr\_mult: 0 \# 学习率为0，其他参数可以看caffe.proto里面的ParamSpec这个类型  
    \}  
    param \{ \# 对应第2个参数blob的配置，也就是全连接层的偏置项的配置  
        lr\_mult: 0 \# 学习率为0  
    \}  
\}

**caffe关闭日志输出消息**

可能对于算法实验阶段来说，我们需要caffe运行网络的时候，打印网络的相关参数消息。然而对于已经走向工程之路的caffe模型来说，我们有可能需要屏蔽那些没用的输出消息，因此这个时候就需要用到caffe的消息屏蔽。

1、在程序最开始运行地方，加入如下代码：

**\[python\]**  [view plain][]   [copy][view plain]

[![在CODE上查看代码片][CODE]][CODE_CODE] [![派生到我的代码片][ico_fork.svg]][ico_fork.svg 1]

1.  ::google::InitGoogleLogging("hjimce");    //初始化     
2.  FLAGS\_stderrthreshold = ::google::ERROR;  //只打印ERROR级别信息

2、在程序结束的时候加入代码：

**\[python\]**  [view plain][]   [copy][view plain]

[![在CODE上查看代码片][CODE]][CODE_CODE] [![派生到我的代码片][ico_fork.svg]][ico_fork.svg 1]

1.  google::ShutdownGoogleLogging();   
2.

[caffe group参数][caffe group]

[Caffe]: http://blog.csdn.net/u010402786/article/details/51262004
[ubuntu14.04_CPU_caffe]: http://blog.csdn.net/samylee/article/details/50935943
[Caffe_ubuntu14.04_Cuda8.0]: http://blog.csdn.net/u012177034/article/details/52102676
[caffe_image_data_layer]: http://blog.csdn.net/kuaitoukid/article/details/51323940
[Caffe_11_db_leveldb_lmdb]: http://www.cnblogs.com/denny402/p/5082341.html
[caffe_leveldb]: http://blog.csdn.net/lingerlanlan/article/details/31361299
[lmdb]: http://blog.csdn.net/sinat_30071459/article/details/50501689
[Python lmdb]: http://blog.csdn.net/ayst123/article/details/44077903
[Mnist_lmdb_leveldb]: http://blog.csdn.net/whiteinblue/article/details/45330801
[Caffe2_cifar10_lmdb_leveldb]: http://blog.csdn.net/whiteinblue/article/details/45367097
[Caffe3_ImageNet_lmdb]: http://blog.csdn.net/whiteinblue/article/details/45485559
[caffe _LMDB]: http://blog.csdn.net/u010668907/article/details/51834411
[Link 1]: http://blog.csdn.net/whiteinblue/article/details/45540301
[Caffe_compute_image_mean]: http://blog.csdn.net/chenriwei2/article/details/46362851
[Caffe_convert_imageset]: http://blog.csdn.net/chenriwei2/article/details/46361333
[train.txt_val.txt]: http://blog.csdn.net/sinat_30071459/article/details/51613304
[txt_lmdb_caffenet]: http://blog.csdn.net/sinat_30071459/article/details/50513263
[Link 2]: http://blog.csdn.net/sinat_30071459/article/details/50519273
[Caffe_test_iter test_interval]: http://blog.csdn.net/iamzhangzhuping/article/details/49993899
[epoch_ iteration_batchsize]: http://blog.csdn.net/sinat_30071459/article/details/50721565
[caffe_BatchNorm]: http://blog.csdn.net/shuzfan/article/details/52729424
[Mnist]: http://blog.csdn.net/u012177034/article/details/52115773
[lenet]: http://blog.csdn.net/u012746763/article/details/50454618
[Caffe 1]: http://blog.csdn.net/u012177034/article/details/52132056
[Link 3]: http://blog.csdn.net/u012177034/article/details/52134205
[Link 4]: http://blog.csdn.net/sinat_30071459/article/details/51723483
[Link 5]: http://blog.csdn.net/sinat_30071459/article/details/51723544
[Link 6]: http://blog.csdn.net/u012177034/article/details/52144325
[Link 7]: http://blog.csdn.net/sinat_30071459/article/details/51724718
[Link 8]: http://blog.csdn.net/u012177034/article/details/52165913
[Blob_Layer_Net]: http://blog.csdn.net/sinat_30071459/article/details/51725099
[Link 9]: http://blog.csdn.net/sinat_30071459/article/details/51773507
[Link 10]: http://blog.csdn.net/langb2014/article/details/50458520
[caffe]: http://caffecn.cn/?/question/123
[caffe_-softmax_loss]: http://blog.csdn.net/shuzfan/article/details/51460895
[caffe_hinge_loss]: http://blog.csdn.net/shuzfan/article/details/53032740
[PairLoss]: http://blog.csdn.net/shuzfan/article/details/51525721
[Windows_VS2013_caffemodel]: http://blog.csdn.net/sinat_30071459/article/details/50974695
[Link 11]: http://blog.csdn.net/sinat_30071459/article/details/51679995
[caffe 1]: http://blog.csdn.net/u012177034/article/category/6357353
[Caffe 2]: http://blog.csdn.net/mmmmmttttff/article/details/51121378
[caffe 2]: http://blog.csdn.net/mmmmmttttff/article/details/51228360
[caffe 3]: http://blog.csdn.net/mmmmmttttff/article/details/51643666
[solver]: http://blog.csdn.net/mmmmmttttff/article/details/51793943
[Caffe 3]: http://www.myexception.cn/other/1828071.html
[Caffe_]: http://blog.csdn.net/chenriwei2/article/details/46300981
[caffe-windows-master]: http://blog.csdn.net/sinat_30071459/article/details/51823390
[Windows_CAFFE_matlab_python]: http://blog.csdn.net/happynear/article/details/45372231
[Caffe 4]: https://github.com/shicai/Caffe_Manual
[What is _lr_policy_ in caffe]: http://blog.csdn.net/yuanchheneducn/article/details/50223299
[caffe_Loss_nan]: http://blog.csdn.net/huangynn/article/details/52947894
[Net Surgery]: http://nbviewer.jupyter.org/github/BVLC/caffe/blob/master/examples/net_surgery.ipynb
[Caffe_loss]: http://caffecn.cn/?/question/43
[fine_tuning]: http://caffecn.cn/?/question/12
[Fine-tuning CaffeNet for Style Recognition on _Flickr Style_ Data]: http://caffe.berkeleyvision.org/gathered/examples/finetune_flickr_style.html
[model_model_model]: http://caffecn.cn/?/question/59
[caffe_model_fineture]: http://www.caffecn.cn/?/question/9
[Caffe _GPU]: http://www.caffecn.cn/?/question/85
[caffe_proto_std]: http://www.caffecn.cn/?/question/37
[global average pooling _average pooling]: http://www.caffecn.cn/?/question/138
[Pooling]: http://www.caffecn.cn/?/question/57
[loss]: http://www.caffecn.cn/?/question/18
[CNN]: http://www.caffecn.cn/?/question/53
[Link 12]: http://caffecn.cn/?/question/4
[End to End]: http://caffecn.cn/?/question/580
[caffe_image_mean]: http://caffecn.cn/?/question/15
[Link 13]: http://caffecn.cn/?/question/535
[state of art _cnn based]: http://caffecn.cn/?/question/154
[caffe_validation_validation]: http://www.caffecn.cn/?/question/736
[Deep]: http://blog.csdn.net/u010751535/article/details/52739803
[Caffe 5]: http://blog.csdn.net/thesby/article/details/50880546
[caffe_ _ caffe.proto]: http://blog.csdn.net/qq_16055159/article/details/45115359
[caffe_Google Protocol Buffers]: http://blog.csdn.net/langb2014/article/details/50151831
[Caffe 6]: http://blog.csdn.net/cheese_pop/article/details/52024980
[Caffe_2_SycedMem]: http://www.cnblogs.com/louyihang-loves-baiyan/p/5150554.html
[caffe_ _ SyncedMemory.cpp]: http://blog.csdn.net/qq_16055159/article/details/45099591
[Caffe 7]: http://blog.csdn.net/thesby/article/details/52682158
[caffe_8_draw_net.py]: http://blog.csdn.net/sinat_30071459/article/details/51773607
[caffe_10LeNet_python]: http://blog.csdn.net/liumaolincycle/article/details/48624131
[caffe_python]: http://blog.csdn.net/sloanqin/article/details/51564582
[pycaffe]: http://blog.csdn.net/thesby/article/details/51218568
[pycaffe 1]: http://blog.csdn.net/thesby/article/details/51259619
[caffe python layer]: http://blog.csdn.net/thesby/article/details/51264439
[caffe_python layer]: http://blog.csdn.net/thesby/article/details/53192104
[caffe python]: http://blog.csdn.net/hjimce/article/details/51884024
[caffe_11]: http://blog.csdn.net/liumaolincycle/article/details/50549974
[caffe_python 1]: http://blog.csdn.net/iamzhangzhuping/article/details/52220442
[caffe_loss]: http://blog.csdn.net/thesby/article/details/51735091
[caffe _4 ----_python_caffe_------by]: http://blog.csdn.net/sloanqin/article/details/49249905
[caffe _3 _caffe_C_------by]: http://blog.csdn.net/sloanqin/article/details/49249279
[Link 14]: http://www.caffecn.cn/?/question/543
[caffe_ ubuntu1404_eclipse]: http://blog.csdn.net/yaoxingfu72/article/details/47999795
[Caffe_LeNet]: http://blog.csdn.net/mounty_fsc/article/details/51090114
[Caffe_LeNet 1]: http://blog.csdn.net/mounty_fsc/article/details/51379395
[Caffe_regression]: http://www.cnblogs.com/frombeijingwithlove/p/5314042.html
[caffe 4]: http://blog.csdn.net/u014114990/article/details/50502125
[cnn]: http://blog.csdn.net/u014114990/article/details/50535742
[caffe 5]: http://blog.csdn.net/lingerlanlan/article/details/37593837
[caffe 6]: http://blog.csdn.net/lingerlanlan/article/details/39400375
[caffe _Crowd counting]: http://blog.csdn.net/seven_first/article/details/52598810
[caffe _ FaceNet _ Triplet Loss Layer]: http://blog.csdn.net/seven_first/article/details/50272005
[caffe _GPU _ _ _GPU]: http://blog.csdn.net/u014114990/article/details/52776990
[caffe adaboost]: http://blog.csdn.net/u014114990/article/details/51005316
[Linux_Caffe]: http://blog.csdn.net/xiaoyezi_1834/article/details/50724875
[train_test.prototxt_deploy.prototxt_slover.prototxt]: http://blog.csdn.net/sunshine_in_moon/article/details/49472901
[view plain]: http://blog.csdn.net/hjimce/article/details/51884024#
[CODE]: https://code.csdn.net/assets/CODE_ico.png
[CODE_CODE]: https://code.csdn.net/snippets/1756875
[ico_fork.svg]: https://code.csdn.net/assets/ico_fork.svg
[ico_fork.svg 1]: https://code.csdn.net/snippets/1756875/fork
[caffe group]: http://blog.csdn.net/liyuan123zhouhui/article/details/70858472