一，train_val.prototxt
name: "CIFAR10_quick"
layer {
name: "cifar"
type: "Data"
top: "data"
top: "label"
include {
phase: TRAIN
}
transform_param {
# mirror: true
# mean_file: "examples/cifar10/mean.binaryproto"uu
mean_file: "myself/00b/00bmean.binaryproto"
}
data_param {
# source: "examples/cifar10/cifar10_train_lmdb"
source: "myself/00b/00b_train_lmdb"
batch_size:
backend: LMDB
}
}
layer {
name: "cifar"
type: "Data"
top: "data"
top: "label"
include {
phase: TEST
}
transform_param {
# mean_file: "examples/cifar10/mean.binaryproto"
mean_file: "myself/00b/00bmean.binaryproto"
}
data_param {
# source: "examples/cifar10/cifar10_test_lmdb"
source: "myself/00b/00b_val_lmdb"
batch_size:
backend: LMDB
}
}
layer {
name: "conv1"
type: "Convolution"
bottom: "data"
top: "conv1"
param {
lr_mult:
}
param {
lr_mult:
}
convolution_param {
num_output:
# pad:
kernel_size:
stride:
weight_filler {
type: "gaussian"
std: 0.0001
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "pool1"
type: "Pooling"
bottom: "conv1"
top: "pool1"
pooling_param {
pool: MAX
kernel_size:
stride:
}
}
layer {
name: "relu1"
type: "ReLU"
bottom: "pool1"
top: "pool1"
}
layer {
name: "conv2"
type: "Convolution"
bottom: "pool1"
top: "conv2"
param {
lr_mult:
}
param {
lr_mult:
}
convolution_param {
num_output:
# pad:
kernel_size:
stride:
weight_filler {
type: "gaussian"
std: 0.01
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "relu2"
type: "ReLU"
bottom: "conv2"
top: "conv2"
}
layer {
name: "pool2"
type: "Pooling"
bottom: "conv2"
top: "pool2"
pooling_param {
pool: AVE
kernel_size:
stride:
}
}
layer {
name: "conv3"
type: "Convolution"
bottom: "pool2"
top: "conv3"
param {
lr_mult:
}
param {
lr_mult:
}
convolution_param {
num_output:
# pad:
kernel_size:
stride:
weight_filler {
type: "gaussian"
std: 0.01
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "relu3"
type: "ReLU"
bottom: "conv3"
top: "conv3"
}
layer {
name: "pool3"
type: "Pooling"
bottom: "conv3"
top: "pool3"
pooling_param {
pool: AVE
kernel_size:
stride:
}
}
layer {
name: "conv4"
type: "Convolution"
bottom: "pool3"
top: "conv4"
param {
lr_mult:
}
param {
lr_mult:
}
convolution_param {
num_output:
# pad:
kernel_size:
stride:
weight_filler {
type: "gaussian"
std: 0.01
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "relu4"
type: "ReLU"
bottom: "conv4"
top: "conv4"
}
layer {
name: "pool4"
type: "Pooling"
bottom: "conv4"
top: "pool4"
pooling_param {
pool: AVE
kernel_size:
stride:
}
}
layer {
name: "ip1"
type: "InnerProduct"
bottom: "pool4"
top: "ip1"
param {
lr_mult:
}
param {
lr_mult:
}
inner_product_param {
num_output:
weight_filler {
type: "gaussian"
std: 0.1
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "ip2"
type: "InnerProduct"
bottom: "ip1"
top: "ip2"
param {
lr_mult:
}
param {
lr_mult:
}
inner_product_param {
num_output:
weight_filler {
type: "gaussian"
std: 0.1
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "accuracy"
type: "Accuracy"
bottom: "ip2"
bottom: "label"
top: "accuracy"
include {
phase: TEST
}
}
layer {
name: "loss"
type: "SoftmaxWithLoss"
bottom: "ip2"
bottom: "label"
top: "loss"
}
二，solver.prototxt
# reduce the learning rate after epochs ( iters) by a factor of # The train/test net protocol buffer definition
net: "myself/00b/train_val.prototxt"
# test_iter specifies how many forward passes the test should carry out.
# In the case of MNIST, we have test batch size and test iterations,
# covering the full , testing images.
test_iter:
# Carry out testing every training iterations.
test_interval:
# The base learning rate, momentum and the weight decay of the network.
base_lr: 0.001
momentum: 0.9
weight_decay: 0.004
# The learning rate policy
lr_policy: "fixed"
# lr_policy: "step"
gamma: 0.1
stepsize:
# Display every iterations
display:
# The maximum number of iterations
max_iter:
# snapshot intermediate results
# snapshot:
# snapshot_format: HDF5
snapshot_prefix: "myself/00b/00b"
# solver mode: CPU or GPU
solver_mode: CPU
三，deploy.prototxt
name: "CIFAR10_quick"
layer {
name: "data"
type: "Input"
top: "data"
input_param { shape: { dim: dim: dim: dim: } }
}
layer {
name: "conv1"
type: "Convolution"
bottom: "data"
top: "conv1"
convolution_param {
num_output:
kernel_size:
stride:
}
}
layer {
name: "relu1"
type: "ReLU"
bottom: "conv1"
top: "conv1"
}
layer {
name: "pool1"
type: "Pooling"
bottom: "conv1"
top: "pool1"
pooling_param {
pool: MAX
kernel_size:
stride:
}
}
layer {
name: "conv2"
type: "Convolution"
bottom: "pool1"
top: "conv2"
convolution_param {
num_output:
kernel_size:
stride:
}
}
layer {
name: "relu2"
type: "ReLU"
bottom: "conv2"
top: "conv2"
}
layer {
name: "pool2"
type: "Pooling"
bottom: "conv2"
top: "pool2"
pooling_param {
pool: MAX
kernel_size:
stride:
}
}
layer {
name: "conv3"
type: "Convolution"
bottom: "pool2"
top: "conv3"
convolution_param {
num_output:
kernel_size:
stride:
}
}
layer {
name: "relu3"
type: "ReLU"
bottom: "conv3"
top: "conv3"
}
layer {
name: "pool3"
type: "Pooling"
bottom: "conv3"
top: "pool3"
pooling_param {
pool: MAX
kernel_size:
stride:
}
}
layer {
name: "conv4"
type: "Convolution"
bottom: "pool3"
top: "conv4"
convolution_param {
num_output:
kernel_size:
stride:
}
}
layer {
name: "relu4"
type: "ReLU"
bottom: "conv4"
top: "conv4"
}
layer {
name: "pool4"
type: "Pooling"
bottom: "conv4"
top: "pool4"
pooling_param {
pool: MAX
kernel_size:
stride:
}
}
layer {
name: "ip1"
type: "InnerProduct"
bottom: "pool4"
top: "ip1"
inner_product_param {
num_output:
}
}
layer {
name: "ip2"
type: "InnerProduct"
bottom: "ip1"
top: "ip2"
inner_product_param {
num_output:
}
}
layer {
#name: "loss" name: "prob"
type: "Softmax"
bottom: "ip2"
top: "prob" #top: "loss"
}
参考一：
模型就用程序自带的caffenet模型，位置在 models/bvlc_reference_caffenet/文件夹下, 将需要的两个配置文件，复制到myfile文件夹内
# sudo cp models/bvlc_reference_caffenet/solver.prototxt examples/myfile/
# sudo cp models/bvlc_reference_caffenet/train_val.prototxt examples/myfile/
修改train_val.protxt，只需要修改两个阶段的data层就可以了，其它可以不用管。
name: "CaffeNet"
layer {
name: "data"
type: "Data"
top: "data"
top: "label"
include {
phase: TRAIN
}
transform_param {
mirror: true
crop_size: 227
mean_file: "examples/myfile/mean.binaryproto"
}
data_param {
source: "examples/myfile/img_train_lmdb"
batch_size: 256
backend: LMDB
}
}
layer {
name: "data"
type: "Data"
top: "data"
top: "label"
include {
phase: TEST
}
transform_param {
mirror: false
crop_size: 227
mean_file: "examples/myfile/mean.binaryproto"
}
data_param {
source: "examples/myfile/img_test_lmdb"
batch_size: 50
backend: LMDB
}
}
实际上就是修改两个data layer的mean_file和source这两个地方，其它都没有变化 。
修改其中的solver.prototxt
# sudo vi examples/myfile/solver.prototxt
net: "examples/myfile/train_val.prototxt"
test_iter: 2
test_interval: 50
base_lr: 0.001
lr_policy: "step"
gamma: 0.1
stepsize: 100
display: 20
max_iter: 500
momentum: 0.9
weight_decay: 0.005
solver_mode: GPU
100个测试数据，batch_size为50，因此test_iter设置为2，就能全cover了。在训练过程中，调整学习率，逐步变小。
参考二：
前面做好了lmdb和均值文件，下面以Googlenet为例修改网络并训练模型。
我们将caffe-master\models下的bvlc_googlenet文件夹复制到caffe-master\examples\imagenet下。（因为我们的lmdb和均值都在这里，放一起方便些）
打开train_val.txt，修改：
1.修改data层：
- layer {
- name: "data"
- type: "Data"
- top: "data"
- top: "label"
- include {
- phase: TRAIN
- }
- transform_param {
- mirror: true
- crop_size: 224
- mean_file: "examples/imagenet/mydata_mean.binaryproto" #均值文件
- #mean_value: 104 #这些注释掉
- #mean_value: 117
- #mean_value: 123
- }
- data_param {
- source: "examples/imagenet/mydata_train_lmdb" #训练集的lmdb
- batch_size: 32 #根据GPU修改
- backend: LMDB
- }
- }
- layer {
- name: "data"
- type: "Data"
- top: "data"
- top: "label"
- include {
- phase: TEST
- }
- transform_param {
- mirror: false
- crop_size: 224
- mean_file: "examples/imagenet/mydata_mean.binaryproto" #均值文件
- #mean_value: 104
- #mean_value: 117
- #mean_value: 123
- }
- data_param {
- source: "examples/imagenet/mydata_val_lmdb" #验证集lmdb
- batch_size: 50 #和solver中的test_iter相乘约等于验证集大小
- backend: LMDB
- }
- }
2.修改输出：
由于Googlenet有三个输出，所以改三个地方，其他网络一般只有一个输出，则改一个地方即可。
如果是微调，那么输出层的层名也要修改。（参数根据层名来初始化，由于输出改了，该层参数就不对应了，因此要改名）
layer {
name: "loss1/classifier"
type: "InnerProduct"
bottom: "loss1/fc"
top: "loss1/classifier"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
inner_product_param {
num_output: 1000 #改成你的数据集类别数
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0
}
}
}
layer {
name: "loss2/classifier"
type: "InnerProduct"
bottom: "loss2/fc"
top: "loss2/classifier"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
inner_product_param {
num_output: 1000 #改成你的数据集类别数
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0
}
}
}
layer {
name: "loss3/classifier"
type: "InnerProduct"
bottom: "pool5/7x7_s1"
top: "loss3/classifier"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
inner_product_param {
num_output: 1000 #改成你的数据集类别数
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0
}
}
}
3.打开deploy.prototxt，修改：
layer {
name: "loss3/classifier"
type: "InnerProduct"
bottom: "pool5/7x7_s1"
top: "loss3/classifier"
param {
lr_mult: 1
decay_mult: 1
}
param {
lr_mult: 2
decay_mult: 0
}
inner_product_param {
num_output: 1000 #改成你的数据集类别数
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
value: 0
}
}
}
如果是微调，该层层名和train_val.prototxt修改一致。
接着，打开solver，修改：
net: "examples/imagenet/bvlc_googlenet/train_val.prototxt" #路径不要错
test_iter: 1000 #前面已说明该值
test_interval: 4000 #迭代多少次测试一次
test_initialization: false
display: 40
average_loss: 40
base_lr: 0.01
lr_policy: "step"
stepsize: 320000 #迭代多少次改变一次学习率
gamma: 0.96
max_iter: 10000000 #迭代次数
momentum: 0.9
weight_decay: 0.0002
snapshot: 40000
snapshot_prefix: "examples/imagenet/bvlc_googlenet" #生成的caffemodel保存在imagenet下，形如bvlc_googlenet_iter_***.caffemodel
solver_mode: GPU
这时，我们回到caffe-master\examples\imagenet下，打开train_caffenet.sh，修改：
(如果是微调，在脚本里加入-weights **/**/**.caffemodel即可，即用来微调的caffemodel路径)
#!/usr/bin/env sh ./build/tools/caffe train \
-solver examples/imagenet/bvlc_googlenet/solver.prototxt -gpu 0
（如果有多个GPU，可自行选择） 然后，在caffe-master下执行改脚本即可开始训练:$caffe-master ./examples/imagenet/train_caffenet.sh
训练得到的caffemodel就可以用来做图像分类了，此时，需要（1）得到的labels.txt，（2）得到的mydata_mean.binaryproto，（3）得到的caffemodel以及已经修改过的deploy.prototxt，共四个文件，具体过程看：http://blog.csdn.net/sinat_30071459/article/details/50974695
参考三：
*_train_test.prototxt,*_deploy.prototxt,*_slover.prototxt文件编写时注意
1、*_train_test.prototxt文件
这是训练与测试网络配置文件
（1）在数据层中 参数include{
phase：TRAIN/TEST
}
TRAIN与TEST不能有“...”否则会报错，还好提示信息里，会提示哪一行出现了问题，如下图：
数字8就代表配置文件的第8行出现了错误
（2）卷积层和全连接层相似：卷积层（Convolution），全连接层（InnerProduct，容易翻译成内积层）相似处有两个【1】：都有两个param{lr_mult：1
decay_mult：1
}
param{lr_mult： 2
decay_mult: 0
}
【2】：convolution_param{}与inner_product_param{}里面的参数相似，甚至相同
今天有事，明天再续！
续上！
（3）平均值文件*_mean.binaryproto要放在transform_param{}里，训练与测试数据集放在data_param{}里
2.*_deploy.prototxt文件
【1】*_deploy.prototxt文件的构造和*_train_test.prototxt文件的构造稍有不同首先没有test网络中的test模块，只有训练模块
【2】数据层的写法和原来也有不同，更加简洁：
input: "data" input_dim: 1 input_dim: 3 input_dim: 32 input_dim: 32
注意红色部分，那是数据层的名字，没有这个的话，第一卷积层无法找到数据，我一开始没有加这句就报错。下面的四个参数有点类似batch_size（1,3,32,32）里四个参数
【3】卷积层和全连接层中weight_filler{}与bias_filler{}两个参数不用再填写，应为这两个参数的值，由已经训练好的模型*.caffemodel文件提供
【4】输出层的变化（1）没有了test模块测试精度（2）输出层
*_train_test.prototxt文件：
layer{ name: "loss" type: "SoftmaxWithLoss"#注意此处与下面的不同 bottom: "ip2" bottom: "label"#注意标签项在下面没有了，因为下面的预测属于哪个标签，因此不能提供标签 top: "loss" }
*_deploy.prototxt文件：
layer { name: "prob" type: "Softmax" bottom: "ip2" top: "prob" }
***注意在两个文件中输出层的类型都发生了变化一个是SoftmaxWithLoss，另一个是Softmax。另外为了方便区分训练与应用输出，训练是输出时是loss，应用时是prob。
3、*_slover.prototxt
net: "test.prototxt" #训练网络的配置文件 test_iter: 100 #test_iter 指明在测试阶段有多上个前向过程（也就是有多少图片）被执行。 在MNIST例子里，在网络配置文件里已经设置test网络的batch size=100，这里test_iter 设置为100，那在测试阶段共有100*100=10000 图片被处理 test_interval: 500 #每500次训练迭代后，执行一次test base_lr: 0.01 #学习率初始化为0.01 momentum：0.9 #u=0.9 weight_decay:0.0005 # lr_policy: "inv" gamma: 0.0001 power: 0.75 #以上三个参数都和降低学习率有关，详细的学习策略和计算公式见下面 // The learning rate decay policy. The currently implemented learning rate
// policies are as follows:
// - fixed: always return base_lr.
// - step: return base_lr * gamma ^ (floor(iter / step))
// - exp: return base_lr * gamma ^ iter
// - inv: return base_lr * (1 + gamma * iter) ^ (- power)
// - multistep: similar to step but it allows non uniform steps defined by
// stepvalue
// - poly: the effective learning rate follows a polynomial decay, to be
// zero by the max_iter. return base_lr (1 - iter/max_iter) ^ (power)
// - sigmoid: the effective learning rate follows a sigmod decay
// return base_lr ( 1/(1 + exp(-gamma * (iter - stepsize))))
// where base_lr, max_iter, gamma, step, stepvalue and power are defined
// in the solver parameter protocol buffer, and iter is the current iteration. display:100 #每100次迭代，显示结果 snapshot: 5000 #每5000次迭代，保存一次快照 snapshot_prefix: "path_prefix" #快照保存前缀：更准确的说是快照保存路径+前缀，应为文件名后的名字是固定的 solver_mode:GPU #选择解算器是用cpu还是gpu
批处理文件编写：
F:/caffe/caffe-windows-master/bin/caffe.exe train --solver=C:/Users/Administrator/Desktop/caffe_test/cifar-10/cifar10_slover_prototxt --gpu=all pause
参考四：
将train_val.prototxt 转换成deploy.prototxt
分类：caffe
（282） （0）
1.删除输入数据（如：type:data...inckude{phase: TRAIN}），然后添加一个数据维度描述。
- input: "data"
- input_dim: 1
- input_dim: 3
- input_dim: 224
- input_dim: 224
- force_backward: true
input: "data"
input_dim: 1
input_dim: 3
input_dim: 224
input_dim: 224
force_backward: true
2.移除最后的“loss” 和“accuracy” 层，加入“prob”层。
- layers {
- name: "prob"
- type: SOFTMAX
- bottom: "fc8"
- top: "prob"
- }
layers {
name: "prob"
type: SOFTMAX
bottom: "fc8"
top: "prob"
}
如果train_val文件中还有其他的预处理层，就稍微复杂点。如下，在'data'层，在‘data’层和‘conv1’层(with bottom:”data” / top:”conv1″). 插入一个层来计算输入数据的均值。
- layer {
- name: “mean”
- type: “Convolution”
- <strong>bottom: “data”
- top: “data”</strong>
- param {
- lr_mult: 0
- decay_mult: 0
- }
- …}
在deploy.prototxt文件中，“mean” 层必须保留，只是容器改变，相应的‘conv1’也要改变 ( bottom:”mean”/ top:”conv1″ )。
- layer {
- name: “mean”
- type: “Convolution”
- <strong>bottom: “data”
- top: “mean“</strong>
- param {
- lr_mult: 0
- decay_mult: 0
- }
- …}