Caffe中Mnist的训练日志解读
2016-02-23 16:55
381 查看
Lenet的网络结构图如下：
Mnist的训练日志如下（其中max_iter设为3，使用CPU模式）
I0223 14:55:42.870805 6406 caffe.cpp:178] Use CPU.//使用CPU模式
I0223 14:55:42.871371 6406 solver.cpp:48] Initializing solver from parameters: //利用solver.prototxt文件，初始化solver参数
test_iter: 100
test_interval: 500
base_lr: 0.01
display: 100
max_iter: 3
lr_policy: "inv"
gamma: 0.0001
power: 0.75
momentum: 0.9
weight_decay: 0.0005
snapshot: 5000
snapshot_prefix: "examples/mnist/lenet"
solver_mode: CPU
net: "examples/mnist/lenet_train_test.prototxt"
I0223 14:55:42.871707 6406 solver.cpp:91] Creating training net from net file: examples/mnist/lenet_train_test.prototxt
I0223 14:55:42.872799 6406 net.cpp:322] The NetState phase (0) differed from the phase (1) specified by a rule in layer mnist
I0223 14:55:42.872871 6406 net.cpp:322] The NetState phase (0) differed from the phase (1) specified by a rule in layer accuracy
I0223 14:55:42.873092 6406 net.cpp:49] Initializing net from parameters: //初始化网络参数
name: "LeNet"
state {
phase: TRAIN
}
layer {
name: "mnist"
type: "Data"
top: "data"
top: "label"
include {
phase: TRAIN
}
transform_param {
scale: 0.00390625
}
data_param {
source: "examples/mnist/mnist_train_lmdb"
batch_size: 64
backend: LMDB
}
}
layer {
name: "conv1"
type: "Convolution"
bottom: "data"
top: "conv1"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
convolution_param {
num_output: 20
kernel_size: 5
stride: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "pool1"
type: "Pooling"
bottom: "conv1"
top: "pool1"
pooling_param {
pool: MAX
kernel_size: 2
stride: 2
}
}
layer {
name: "conv2"
type: "Convolution"
bottom: "pool1"
top: "conv2"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
convolution_param {
num_output: 50
kernel_size: 5
stride: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "pool2"
type: "Pooling"
bottom: "conv2"
top: "pool2"
pooling_param {
pool: MAX
kernel_size: 2
stride: 2
}
}
layer {
name: "ip1"
type: "InnerProduct"
bottom: "pool2"
top: "ip1"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
inner_product_param {
num_output: 500
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "relu1"
type: "ReLU"
bottom: "ip1"
top: "ip1"
}
layer {
name: "ip2"
type: "InnerProduct"
bottom: "ip1"
top: "ip2"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
inner_product_param {
num_output: 10
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "loss"
type: "SoftmaxWithLoss"
bottom: "ip2"
bottom: "label"
top: "loss"
}
//构造Lenet网络结构
I0223 14:55:42.873805 6406 layer_factory.hpp:77] Creating layer mnist//构建mnist层，也就是数据层
I0223 14:55:42.875214 6406 net.cpp:106] Creating Layer mnist
I0223 14:55:42.875300 6406 net.cpp:411] mnist -> data
I0223 14:55:42.875455 6409 db_lmdb.cpp:38] Opened lmdb examples/mnist/mnist_train_lmdb//使用lmdb格式的训练数据
I0223 14:55:42.875653 6406 net.cpp:411] mnist -> label
//数据的输出大小为64*1*28*28(64：图像数，此处为batch大小 1：通道数 28：图像的高 28：图像的宽)，数据层只有输出没有输入
I0223 14:55:42.875879 6406 data_layer.cpp:41] output data size: 64,1,28,28
I0223 14:55:42.876302 6406 base_data_layer.cpp:69] Initializing prefetch//初始化prefetch，该函数可以提前获取下一个batch，提高效率
I0223 14:55:42.876423 6406 base_data_layer.cpp:72] Prefetch initialized.
I0223 14:55:42.876461 6406 net.cpp:150] Setting up mnist
I0223 14:55:42.876531 6406 net.cpp:157] Top shape: 64 1 28 28 (50176)//50176=64*1*28*28
I0223 14:55:42.876636 6406 net.cpp:157] Top shape: 64 (64)
I0223 14:55:42.876716 6406 net.cpp:165] Memory required for data: 200960//此时所需要的内存大小
I0223 14:55:42.876829 6406 layer_factory.hpp:77] Creating layer conv1//构建卷积层conv1
I0223 14:55:42.877009 6406 net.cpp:106] Creating Layer conv1
I0223 14:55:42.877122 6406 net.cpp:454] conv1 <- data//数据层的top为卷积层conv1（即数据层data的后面一层为conv1）
I0223 14:55:42.877312 6406 net.cpp:411] conv1 -> conv1
I0223 14:55:42.877905 6406 net.cpp:150] Setting up conv1
//conv1的输出数据大小（20为通道数，也是卷积核的个数），卷积核大小为5，步长为1，经过卷积后图像大小已由28*28变为24*24
I0223 14:55:42.877964 6406 net.cpp:157] Top shape: 64 20 24 24 (737280)
I0223 14:55:42.877993 6406 net.cpp:165] Memory required for data: 3150080
I0223 14:55:42.878154 6406 layer_factory.hpp:77] Creating layer pool1//构建池化层pool1（使用的是最大池化方式）
I0223 14:55:42.878242 6406 net.cpp:106] Creating Layer pool1
I0223 14:55:42.878325 6406 net.cpp:454] pool1 <- conv1//卷积层conv1的top为池化层pool1
I0223 14:55:42.878412 6406 net.cpp:411] pool1 -> pool1
I0223 14:55:42.878540 6406 net.cpp:150] Setting up pool1
I0223 14:55:42.878597 6406 net.cpp:157] Top shape: 64 20 12 12 (184320)//经池化后，图像大小已由24*24变成12*12，池化大小为2*2
I0223 14:55:42.878628 6406 net.cpp:165] Memory required for data: 3887360
I0223 14:55:42.878662 6406 layer_factory.hpp:77] Creating layer conv2//构建卷积层conv2
I0223 14:55:42.878742 6406 net.cpp:106] Creating Layer conv2
I0223 14:55:42.878779 6406 net.cpp:454] conv2 <- pool1//池化层pool1的top为卷积层conv2
I0223 14:55:42.878862 6406 net.cpp:411] conv2 -> conv2
I0223 14:55:42.882365 6410 data_layer.cpp:102] Prefetch batch: 5 ms.//提前获取下一个batch
I0223 14:55:42.882421 6410 data_layer.cpp:103] Read time: 0.791 ms.//batch的读取时间
I0223 14:55:42.882447 6410 data_layer.cpp:104] Transform time: 3.296 ms.//batch的转化时间
I0223 14:55:42.883898 6406 net.cpp:150] Setting up conv2//配置卷积层conv2
//卷积层conv2的输出大小由64*20*12*12变为64*50*8*8，卷积核数目为50，卷积核大小为5，步长为1
I0223 14:55:42.883935 6406 net.cpp:157] Top shape: 64 50 8 8 (204800)
I0223 14:55:42.883962 6406 net.cpp:165] Memory required for data: 4706560
I0223 14:55:42.884052 6406 layer_factory.hpp:77] Creating layer pool2//构建池化层pool2
I0223 14:55:42.884110 6406 net.cpp:106] Creating Layer pool2
I0223 14:55:42.884181 6406 net.cpp:454] pool2 <- conv2//卷积层conv2的top为池化层pool2
I0223 14:55:42.884268 6406 net.cpp:411] pool2 -> pool2
I0223 14:55:42.884383 6406 net.cpp:150] Setting up pool2
I0223 14:55:42.884448 6406 net.cpp:157] Top shape: 64 50 4 4 (51200)//经池化后，图像大小已由8*8变成4*4，池化大小为2*2
I0223 14:55:42.884500 6406 net.cpp:165] Memory required for data: 4911360
I0223 14:55:42.884553 6406 layer_factory.hpp:77] Creating layer ip1//构建内积层ip1（InnerProduct）
I0223 14:55:42.884644 6406 net.cpp:106] Creating Layer ip1
I0223 14:55:42.884706 6406 net.cpp:454] ip1 <- pool2//池化层pool2的top为内积层ip1
I0223 14:55:42.884790 6406 net.cpp:411] ip1 -> ip1
I0223 14:55:42.887511 6410 data_layer.cpp:102] Prefetch batch: 4 ms.
I0223 14:55:42.887596 6410 data_layer.cpp:103] Read time: 0.887 ms.
I0223 14:55:42.887634 6410 data_layer.cpp:104] Transform time: 2.494 ms.
I0223 14:55:42.892683 6410 data_layer.cpp:102] Prefetch batch: 4 ms.
I0223 14:55:42.892719 6410 data_layer.cpp:103] Read time: 0.536 ms.
I0223 14:55:42.892736 6410 data_layer.cpp:104] Transform time: 3.193 ms.
I0223 14:55:42.929069 6406 net.cpp:150] Setting up ip1//配置内积层ip1
I0223 14:55:42.929119 6406 net.cpp:157] Top shape: 64 500 (32000)//数据的输出大小为64*500
I0223 14:55:42.929129 6406 net.cpp:165] Memory required for data: 5039360
I0223 14:55:42.929193 6406 layer_factory.hpp:77] Creating layer relu1//构建非线性变换层relu1
I0223 14:55:42.929252 6406 net.cpp:106] Creating Layer relu1
I0223 14:55:42.929280 6406 net.cpp:454] relu1 <- ip1//内积层ip1的数据传送至非线性变换层relu1
I0223 14:55:42.929325 6406 net.cpp:397] relu1 -> ip1 (in-place)//数据经过非线性变换后传回内积层ip1
I0223 14:55:42.929373 6406 net.cpp:150] Setting up relu1
I0223 14:55:42.929394 6406 net.cpp:157] Top shape: 64 500 (32000)//数据的输出大小为64*500
I0223 14:55:42.929409 6406 net.cpp:165] Memory required for data: 5167360
I0223 14:55:42.929425 6406 layer_factory.hpp:77] Creating layer ip2//构建内积层ip2
I0223 14:55:42.929464 6406 net.cpp:106] Creating Layer ip2
I0223 14:55:42.929483 6406 net.cpp:454] ip2 <- ip1//内积层ip1的top为ip2
I0223 14:55:42.929522 6406 net.cpp:411] ip2 -> ip2
I0223 14:55:42.930277 6406 net.cpp:150] Setting up ip2
I0223 14:55:42.930307 6406 net.cpp:157] Top shape: 64 10 (640)//数据的输出大小为64*10
I0223 14:55:42.930325 6406 net.cpp:165] Memory required for data: 5169920
I0223 14:55:42.930368 6406 layer_factory.hpp:77] Creating layer loss
I0223 14:55:42.930424 6406 net.cpp:106] Creating Layer loss//构建loss层loss
I0223 14:55:42.930449 6406 net.cpp:454] loss <- ip2//内积层ip2的top为loss层
I0223 14:55:42.930486 6406 net.cpp:454] loss <- label//数据标签层label的top为loss层
I0223 14:55:42.930532 6406 net.cpp:411] loss -> loss//利用Softmax计算loss
I0223 14:55:42.930603 6406 layer_factory.hpp:77] Creating layer loss
I0223 14:55:42.930709 6406 net.cpp:150] Setting up loss
I0223 14:55:42.930735 6406 net.cpp:157] Top shape: (1)//loss层的数据输出大小为1
I0223 14:55:42.930747 6406 net.cpp:160] with loss weight 1//loss的权重为1
I0223 14:55:42.930768 6406 net.cpp:165] Memory required for data: 5169924
I0223 14:55:42.930788 6406 net.cpp:226] loss needs backward computation.//进行反向传播梯度
I0223 14:55:42.930810 6406 net.cpp:226] ip2 needs backward computation.
I0223 14:55:42.930827 6406 net.cpp:226] relu1 needs backward computation.
I0223 14:55:42.930843 6406 net.cpp:226] ip1 needs backward computation.
I0223 14:55:42.930887 6406 net.cpp:226] pool2 needs backward computation.
I0223 14:55:42.930909 6406 net.cpp:226] conv2 needs backward computation.
I0223 14:55:42.930930 6406 net.cpp:226] pool1 needs backward computation.
I0223 14:55:42.930951 6406 net.cpp:226] conv1 needs backward computation.
I0223 14:55:42.930973 6406 net.cpp:228] mnist does not need backward computation.
I0223 14:55:42.930984 6406 net.cpp:270] This network produces output loss
I0223 14:55:42.931020 6406 net.cpp:283] Network initialization done.
//接下来是进行测试网络的构建与测试过程，总共100次迭代过程，每次迭代的batch大小为1000，故测试集的大小为100*100=10000
I0223 14:55:42.931638 6406 solver.cpp:181] Creating test net (#0) specified by net file: examples/mnist/lenet_train_test.prototxt
I0223 14:55:42.931759 6406 net.cpp:322] The NetState phase (1) differed from the phase (0) specified by a rule in layer mnist
I0223 14:55:42.931941 6406 net.cpp:49] Initializing net from parameters:
name: "LeNet"
state {
phase: TEST
}
layer {
name: "mnist"
type: "Data"
top: "data"
top: "label"
include {
phase: TEST
}
transform_param {
scale: 0.00390625
}
data_param {
source: "examples/mnist/mnist_test_lmdb"
batch_size: 100
backend: LMDB
}
}
layer {
name: "conv1"
type: "Convolution"
bottom: "data"
top: "conv1"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
convolution_param {
num_output: 20
kernel_size: 5
stride: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "pool1"
type: "Pooling"
bottom: "conv1"
top: "pool1"
pooling_param {
pool: MAX
kernel_size: 2
stride: 2
}
}
layer {
name: "conv2"
type: "Convolution"
bottom: "pool1"
top: "conv2"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
convolution_param {
num_output: 50
kernel_size: 5
stride: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "pool2"
type: "Pooling"
bottom: "conv2"
top: "pool2"
pooling_param {
pool: MAX
kernel_size: 2
stride: 2
}
}
layer {
name: "ip1"
type: "InnerProduct"
bottom: "pool2"
top: "ip1"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
inner_product_param {
num_output: 500
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "relu1"
type: "ReLU"
bottom: "ip1"
top: "ip1"
}
layer {
name: "ip2"
type: "InnerProduct"
bottom: "ip1"
top: "ip2"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
inner_product_param {
num_output: 10
weight_filler {
type: "xavier"
}
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"
}
I0223 14:55:42.932307 6406 layer_factory.hpp:77] Creating layer mnist
I0223 14:55:42.934370 6406 net.cpp:106] Creating Layer mnist
I0223 14:55:42.934464 6406 net.cpp:411] mnist -> data
I0223 14:55:42.934545 6411 db_lmdb.cpp:38] Opened lmdb examples/mnist/mnist_test_lmdb
I0223 14:55:42.934552 6406 net.cpp:411] mnist -> label
I0223 14:55:42.934733 6406 data_layer.cpp:41] output data size: 100,1,28,28
I0223 14:55:42.935247 6406 base_data_layer.cpp:69] Initializing prefetch
I0223 14:55:42.935431 6406 base_data_layer.cpp:72] Prefetch initialized.
I0223 14:55:42.935503 6406 net.cpp:150] Setting up mnist
I0223 14:55:42.935598 6406 net.cpp:157] Top shape: 100 1 28 28 (78400)
I0223 14:55:42.935681 6406 net.cpp:157] Top shape: 100 (100)
I0223 14:55:42.935739 6406 net.cpp:165] Memory required for data: 314000
I0223 14:55:42.935811 6406 layer_factory.hpp:77] Creating layer label_mnist_1_split
I0223 14:55:42.935909 6406 net.cpp:106] Creating Layer label_mnist_1_split
I0223 14:55:42.935940 6406 net.cpp:454] label_mnist_1_split <- label
I0223 14:55:42.935992 6406 net.cpp:411] label_mnist_1_split -> label_mnist_1_split_0
I0223 14:55:42.936048 6406 net.cpp:411] label_mnist_1_split -> label_mnist_1_split_1
I0223 14:55:42.936138 6406 net.cpp:150] Setting up label_mnist_1_split
I0223 14:55:42.936182 6406 net.cpp:157] Top shape: 100 (100)
I0223 14:55:42.936202 6406 net.cpp:157] Top shape: 100 (100)
I0223 14:55:42.936223 6406 net.cpp:165] Memory required for data: 314800
I0223 14:55:42.936246 6406 layer_factory.hpp:77] Creating layer conv1
I0223 14:55:42.936327 6406 net.cpp:106] Creating Layer conv1
I0223 14:55:42.936350 6406 net.cpp:454] conv1 <- data
I0223 14:55:42.936396 6406 net.cpp:411] conv1 -> conv1
I0223 14:55:42.936653 6406 net.cpp:150] Setting up conv1
I0223 14:55:42.936681 6406 net.cpp:157] Top shape: 100 20 24 24 (1152000)
I0223 14:55:42.936693 6406 net.cpp:165] Memory required for data: 4922800
I0223 14:55:42.936758 6406 layer_factory.hpp:77] Creating layer pool1
I0223 14:55:42.936802 6406 net.cpp:106] Creating Layer pool1
I0223 14:55:42.936823 6406 net.cpp:454] pool1 <- conv1
I0223 14:55:42.936872 6406 net.cpp:411] pool1 -> pool1
I0223 14:55:42.936938 6406 net.cpp:150] Setting up pool1
I0223 14:55:42.936962 6406 net.cpp:157] Top shape: 100 20 12 12 (288000)
I0223 14:55:42.936975 6406 net.cpp:165] Memory required for data: 6074800
I0223 14:55:42.936990 6406 layer_factory.hpp:77] Creating layer conv2
I0223 14:55:42.937044 6406 net.cpp:106] Creating Layer conv2
I0223 14:55:42.937064 6406 net.cpp:454] conv2 <- pool1
I0223 14:55:42.937110 6406 net.cpp:411] conv2 -> conv2
I0223 14:55:42.940639 6406 net.cpp:150] Setting up conv2
I0223 14:55:42.940696 6406 net.cpp:157] Top shape: 100 50 8 8 (320000)
I0223 14:55:42.940711 6406 net.cpp:165] Memory required for data: 7354800
I0223 14:55:42.940803 6406 layer_factory.hpp:77] Creating layer pool2
I0223 14:55:42.940876 6406 net.cpp:106] Creating Layer pool2
I0223 14:55:42.940910 6406 net.cpp:454] pool2 <- conv2
I0223 14:55:42.940970 6406 net.cpp:411] pool2 -> pool2
I0223 14:55:42.941061 6406 net.cpp:150] Setting up pool2
I0223 14:55:42.941093 6406 net.cpp:157] Top shape: 100 50 4 4 (80000)
I0223 14:55:42.941108 6406 net.cpp:165] Memory required for data: 7674800
I0223 14:55:42.941131 6406 layer_factory.hpp:77] Creating layer ip1
I0223 14:55:42.941184 6406 net.cpp:106] Creating Layer ip1
I0223 14:55:42.941210 6406 net.cpp:454] ip1 <- pool2
I0223 14:55:42.941262 6406 net.cpp:411] ip1 -> ip1
……
//输出准确率accuracy和损失loss
I0223 14:55:59.623518 6406 solver.cpp:408] Test net output #0: accuracy = 0.1309
I0223 14:55:59.623611 6406 solver.cpp:408] Test net output #1: loss = 2.31399 (* 1 = 2.31399 loss)
I0223 14:55:59.623705 6406 base_data_layer.cpp:115] Prefetch copied
I0223 14:55:59.625409 6410 data_layer.cpp:102] Prefetch batch: 1 ms.
I0223 14:55:59.625458 6410 data_layer.cpp:103] Read time: 0.188 ms.
I0223 14:55:59.625468 6410 data_layer.cpp:104] Transform time: 1.022 ms.
I0223 14:55:59.874915 6406 solver.cpp:229] Iteration 0, loss = 2.30499
I0223 14:55:59.874987 6406 solver.cpp:245] Train net output #0: loss = 2.30499 (* 1 = 2.30499 loss)
I0223 14:55:59.875026 6406 sgd_solver.cpp:106] Iteration 0, lr = 0.01
I0223 14:55:59.879196 6406 base_data_layer.cpp:115] Prefetch copied
I0223 14:55:59.881901 6410 data_layer.cpp:102] Prefetch batch: 2 ms.
I0223 14:55:59.881934 6410 data_layer.cpp:103] Read time: 0.413 ms.
I0223 14:55:59.881945 6410 data_layer.cpp:104] Transform time: 1.312 ms.
I0223 14:56:00.149636 6406 base_data_layer.cpp:115] Prefetch copied
I0223 14:56:00.151485 6410 data_layer.cpp:102] Prefetch batch: 1 ms.
I0223 14:56:00.151515 6410 data_layer.cpp:103] Read time: 0.195 ms.
I0223 14:56:00.151525 6410 data_layer.cpp:104] Transform time: 1.073 ms.
I0223 14:56:00.400404 6406 solver.cpp:458] Snapshotting to binary proto file examples/mnist/lenet_iter_3.caffemodel//保存当前训练模型快照
I0223 14:56:00.400461 6406 net.cpp:918] Serializing 9 layers
//保存当前solver的快照
I0223 14:56:00.434458 6406 sgd_solver.cpp:273] Snapshotting solver state to binary proto file examples/mnist/lenet_iter_3.solverstate
I0223 14:56:00.455166 6406 solver.cpp:323] Optimization Done.
I0223 14:56:00.455202 6406 caffe.cpp:222] Optimization Done.//程序运行结束
Mnist的训练日志如下（其中max_iter设为3，使用CPU模式）
I0223 14:55:42.870805 6406 caffe.cpp:178] Use CPU.//使用CPU模式
I0223 14:55:42.871371 6406 solver.cpp:48] Initializing solver from parameters: //利用solver.prototxt文件，初始化solver参数
test_iter: 100
test_interval: 500
base_lr: 0.01
display: 100
max_iter: 3
lr_policy: "inv"
gamma: 0.0001
power: 0.75
momentum: 0.9
weight_decay: 0.0005
snapshot: 5000
snapshot_prefix: "examples/mnist/lenet"
solver_mode: CPU
net: "examples/mnist/lenet_train_test.prototxt"
I0223 14:55:42.871707 6406 solver.cpp:91] Creating training net from net file: examples/mnist/lenet_train_test.prototxt
I0223 14:55:42.872799 6406 net.cpp:322] The NetState phase (0) differed from the phase (1) specified by a rule in layer mnist
I0223 14:55:42.872871 6406 net.cpp:322] The NetState phase (0) differed from the phase (1) specified by a rule in layer accuracy
I0223 14:55:42.873092 6406 net.cpp:49] Initializing net from parameters: //初始化网络参数
name: "LeNet"
state {
phase: TRAIN
}
layer {
name: "mnist"
type: "Data"
top: "data"
top: "label"
include {
phase: TRAIN
}
transform_param {
scale: 0.00390625
}
data_param {
source: "examples/mnist/mnist_train_lmdb"
batch_size: 64
backend: LMDB
}
}
layer {
name: "conv1"
type: "Convolution"
bottom: "data"
top: "conv1"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
convolution_param {
num_output: 20
kernel_size: 5
stride: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "pool1"
type: "Pooling"
bottom: "conv1"
top: "pool1"
pooling_param {
pool: MAX
kernel_size: 2
stride: 2
}
}
layer {
name: "conv2"
type: "Convolution"
bottom: "pool1"
top: "conv2"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
convolution_param {
num_output: 50
kernel_size: 5
stride: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "pool2"
type: "Pooling"
bottom: "conv2"
top: "pool2"
pooling_param {
pool: MAX
kernel_size: 2
stride: 2
}
}
layer {
name: "ip1"
type: "InnerProduct"
bottom: "pool2"
top: "ip1"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
inner_product_param {
num_output: 500
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "relu1"
type: "ReLU"
bottom: "ip1"
top: "ip1"
}
layer {
name: "ip2"
type: "InnerProduct"
bottom: "ip1"
top: "ip2"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
inner_product_param {
num_output: 10
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "loss"
type: "SoftmaxWithLoss"
bottom: "ip2"
bottom: "label"
top: "loss"
}
//构造Lenet网络结构
I0223 14:55:42.873805 6406 layer_factory.hpp:77] Creating layer mnist//构建mnist层，也就是数据层
I0223 14:55:42.875214 6406 net.cpp:106] Creating Layer mnist
I0223 14:55:42.875300 6406 net.cpp:411] mnist -> data
I0223 14:55:42.875455 6409 db_lmdb.cpp:38] Opened lmdb examples/mnist/mnist_train_lmdb//使用lmdb格式的训练数据
I0223 14:55:42.875653 6406 net.cpp:411] mnist -> label
//数据的输出大小为64*1*28*28(64：图像数，此处为batch大小 1：通道数 28：图像的高 28：图像的宽)，数据层只有输出没有输入
I0223 14:55:42.875879 6406 data_layer.cpp:41] output data size: 64,1,28,28
I0223 14:55:42.876302 6406 base_data_layer.cpp:69] Initializing prefetch//初始化prefetch，该函数可以提前获取下一个batch，提高效率
I0223 14:55:42.876423 6406 base_data_layer.cpp:72] Prefetch initialized.
I0223 14:55:42.876461 6406 net.cpp:150] Setting up mnist
I0223 14:55:42.876531 6406 net.cpp:157] Top shape: 64 1 28 28 (50176)//50176=64*1*28*28
I0223 14:55:42.876636 6406 net.cpp:157] Top shape: 64 (64)
I0223 14:55:42.876716 6406 net.cpp:165] Memory required for data: 200960//此时所需要的内存大小
I0223 14:55:42.876829 6406 layer_factory.hpp:77] Creating layer conv1//构建卷积层conv1
I0223 14:55:42.877009 6406 net.cpp:106] Creating Layer conv1
I0223 14:55:42.877122 6406 net.cpp:454] conv1 <- data//数据层的top为卷积层conv1（即数据层data的后面一层为conv1）
I0223 14:55:42.877312 6406 net.cpp:411] conv1 -> conv1
I0223 14:55:42.877905 6406 net.cpp:150] Setting up conv1
//conv1的输出数据大小（20为通道数，也是卷积核的个数），卷积核大小为5，步长为1，经过卷积后图像大小已由28*28变为24*24
I0223 14:55:42.877964 6406 net.cpp:157] Top shape: 64 20 24 24 (737280)
I0223 14:55:42.877993 6406 net.cpp:165] Memory required for data: 3150080
I0223 14:55:42.878154 6406 layer_factory.hpp:77] Creating layer pool1//构建池化层pool1（使用的是最大池化方式）
I0223 14:55:42.878242 6406 net.cpp:106] Creating Layer pool1
I0223 14:55:42.878325 6406 net.cpp:454] pool1 <- conv1//卷积层conv1的top为池化层pool1
I0223 14:55:42.878412 6406 net.cpp:411] pool1 -> pool1
I0223 14:55:42.878540 6406 net.cpp:150] Setting up pool1
I0223 14:55:42.878597 6406 net.cpp:157] Top shape: 64 20 12 12 (184320)//经池化后，图像大小已由24*24变成12*12，池化大小为2*2
I0223 14:55:42.878628 6406 net.cpp:165] Memory required for data: 3887360
I0223 14:55:42.878662 6406 layer_factory.hpp:77] Creating layer conv2//构建卷积层conv2
I0223 14:55:42.878742 6406 net.cpp:106] Creating Layer conv2
I0223 14:55:42.878779 6406 net.cpp:454] conv2 <- pool1//池化层pool1的top为卷积层conv2
I0223 14:55:42.878862 6406 net.cpp:411] conv2 -> conv2
I0223 14:55:42.882365 6410 data_layer.cpp:102] Prefetch batch: 5 ms.//提前获取下一个batch
I0223 14:55:42.882421 6410 data_layer.cpp:103] Read time: 0.791 ms.//batch的读取时间
I0223 14:55:42.882447 6410 data_layer.cpp:104] Transform time: 3.296 ms.//batch的转化时间
I0223 14:55:42.883898 6406 net.cpp:150] Setting up conv2//配置卷积层conv2
//卷积层conv2的输出大小由64*20*12*12变为64*50*8*8，卷积核数目为50，卷积核大小为5，步长为1
I0223 14:55:42.883935 6406 net.cpp:157] Top shape: 64 50 8 8 (204800)
I0223 14:55:42.883962 6406 net.cpp:165] Memory required for data: 4706560
I0223 14:55:42.884052 6406 layer_factory.hpp:77] Creating layer pool2//构建池化层pool2
I0223 14:55:42.884110 6406 net.cpp:106] Creating Layer pool2
I0223 14:55:42.884181 6406 net.cpp:454] pool2 <- conv2//卷积层conv2的top为池化层pool2
I0223 14:55:42.884268 6406 net.cpp:411] pool2 -> pool2
I0223 14:55:42.884383 6406 net.cpp:150] Setting up pool2
I0223 14:55:42.884448 6406 net.cpp:157] Top shape: 64 50 4 4 (51200)//经池化后，图像大小已由8*8变成4*4，池化大小为2*2
I0223 14:55:42.884500 6406 net.cpp:165] Memory required for data: 4911360
I0223 14:55:42.884553 6406 layer_factory.hpp:77] Creating layer ip1//构建内积层ip1（InnerProduct）
I0223 14:55:42.884644 6406 net.cpp:106] Creating Layer ip1
I0223 14:55:42.884706 6406 net.cpp:454] ip1 <- pool2//池化层pool2的top为内积层ip1
I0223 14:55:42.884790 6406 net.cpp:411] ip1 -> ip1
I0223 14:55:42.887511 6410 data_layer.cpp:102] Prefetch batch: 4 ms.
I0223 14:55:42.887596 6410 data_layer.cpp:103] Read time: 0.887 ms.
I0223 14:55:42.887634 6410 data_layer.cpp:104] Transform time: 2.494 ms.
I0223 14:55:42.892683 6410 data_layer.cpp:102] Prefetch batch: 4 ms.
I0223 14:55:42.892719 6410 data_layer.cpp:103] Read time: 0.536 ms.
I0223 14:55:42.892736 6410 data_layer.cpp:104] Transform time: 3.193 ms.
I0223 14:55:42.929069 6406 net.cpp:150] Setting up ip1//配置内积层ip1
I0223 14:55:42.929119 6406 net.cpp:157] Top shape: 64 500 (32000)//数据的输出大小为64*500
I0223 14:55:42.929129 6406 net.cpp:165] Memory required for data: 5039360
I0223 14:55:42.929193 6406 layer_factory.hpp:77] Creating layer relu1//构建非线性变换层relu1
I0223 14:55:42.929252 6406 net.cpp:106] Creating Layer relu1
I0223 14:55:42.929280 6406 net.cpp:454] relu1 <- ip1//内积层ip1的数据传送至非线性变换层relu1
I0223 14:55:42.929325 6406 net.cpp:397] relu1 -> ip1 (in-place)//数据经过非线性变换后传回内积层ip1
I0223 14:55:42.929373 6406 net.cpp:150] Setting up relu1
I0223 14:55:42.929394 6406 net.cpp:157] Top shape: 64 500 (32000)//数据的输出大小为64*500
I0223 14:55:42.929409 6406 net.cpp:165] Memory required for data: 5167360
I0223 14:55:42.929425 6406 layer_factory.hpp:77] Creating layer ip2//构建内积层ip2
I0223 14:55:42.929464 6406 net.cpp:106] Creating Layer ip2
I0223 14:55:42.929483 6406 net.cpp:454] ip2 <- ip1//内积层ip1的top为ip2
I0223 14:55:42.929522 6406 net.cpp:411] ip2 -> ip2
I0223 14:55:42.930277 6406 net.cpp:150] Setting up ip2
I0223 14:55:42.930307 6406 net.cpp:157] Top shape: 64 10 (640)//数据的输出大小为64*10
I0223 14:55:42.930325 6406 net.cpp:165] Memory required for data: 5169920
I0223 14:55:42.930368 6406 layer_factory.hpp:77] Creating layer loss
I0223 14:55:42.930424 6406 net.cpp:106] Creating Layer loss//构建loss层loss
I0223 14:55:42.930449 6406 net.cpp:454] loss <- ip2//内积层ip2的top为loss层
I0223 14:55:42.930486 6406 net.cpp:454] loss <- label//数据标签层label的top为loss层
I0223 14:55:42.930532 6406 net.cpp:411] loss -> loss//利用Softmax计算loss
I0223 14:55:42.930603 6406 layer_factory.hpp:77] Creating layer loss
I0223 14:55:42.930709 6406 net.cpp:150] Setting up loss
I0223 14:55:42.930735 6406 net.cpp:157] Top shape: (1)//loss层的数据输出大小为1
I0223 14:55:42.930747 6406 net.cpp:160] with loss weight 1//loss的权重为1
I0223 14:55:42.930768 6406 net.cpp:165] Memory required for data: 5169924
I0223 14:55:42.930788 6406 net.cpp:226] loss needs backward computation.//进行反向传播梯度
I0223 14:55:42.930810 6406 net.cpp:226] ip2 needs backward computation.
I0223 14:55:42.930827 6406 net.cpp:226] relu1 needs backward computation.
I0223 14:55:42.930843 6406 net.cpp:226] ip1 needs backward computation.
I0223 14:55:42.930887 6406 net.cpp:226] pool2 needs backward computation.
I0223 14:55:42.930909 6406 net.cpp:226] conv2 needs backward computation.
I0223 14:55:42.930930 6406 net.cpp:226] pool1 needs backward computation.
I0223 14:55:42.930951 6406 net.cpp:226] conv1 needs backward computation.
I0223 14:55:42.930973 6406 net.cpp:228] mnist does not need backward computation.
I0223 14:55:42.930984 6406 net.cpp:270] This network produces output loss
I0223 14:55:42.931020 6406 net.cpp:283] Network initialization done.
//接下来是进行测试网络的构建与测试过程，总共100次迭代过程，每次迭代的batch大小为1000，故测试集的大小为100*100=10000
I0223 14:55:42.931638 6406 solver.cpp:181] Creating test net (#0) specified by net file: examples/mnist/lenet_train_test.prototxt
I0223 14:55:42.931759 6406 net.cpp:322] The NetState phase (1) differed from the phase (0) specified by a rule in layer mnist
I0223 14:55:42.931941 6406 net.cpp:49] Initializing net from parameters:
name: "LeNet"
state {
phase: TEST
}
layer {
name: "mnist"
type: "Data"
top: "data"
top: "label"
include {
phase: TEST
}
transform_param {
scale: 0.00390625
}
data_param {
source: "examples/mnist/mnist_test_lmdb"
batch_size: 100
backend: LMDB
}
}
layer {
name: "conv1"
type: "Convolution"
bottom: "data"
top: "conv1"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
convolution_param {
num_output: 20
kernel_size: 5
stride: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "pool1"
type: "Pooling"
bottom: "conv1"
top: "pool1"
pooling_param {
pool: MAX
kernel_size: 2
stride: 2
}
}
layer {
name: "conv2"
type: "Convolution"
bottom: "pool1"
top: "conv2"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
convolution_param {
num_output: 50
kernel_size: 5
stride: 1
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "pool2"
type: "Pooling"
bottom: "conv2"
top: "pool2"
pooling_param {
pool: MAX
kernel_size: 2
stride: 2
}
}
layer {
name: "ip1"
type: "InnerProduct"
bottom: "pool2"
top: "ip1"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
inner_product_param {
num_output: 500
weight_filler {
type: "xavier"
}
bias_filler {
type: "constant"
}
}
}
layer {
name: "relu1"
type: "ReLU"
bottom: "ip1"
top: "ip1"
}
layer {
name: "ip2"
type: "InnerProduct"
bottom: "ip1"
top: "ip2"
param {
lr_mult: 1
}
param {
lr_mult: 2
}
inner_product_param {
num_output: 10
weight_filler {
type: "xavier"
}
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"
}
I0223 14:55:42.932307 6406 layer_factory.hpp:77] Creating layer mnist
I0223 14:55:42.934370 6406 net.cpp:106] Creating Layer mnist
I0223 14:55:42.934464 6406 net.cpp:411] mnist -> data
I0223 14:55:42.934545 6411 db_lmdb.cpp:38] Opened lmdb examples/mnist/mnist_test_lmdb
I0223 14:55:42.934552 6406 net.cpp:411] mnist -> label
I0223 14:55:42.934733 6406 data_layer.cpp:41] output data size: 100,1,28,28
I0223 14:55:42.935247 6406 base_data_layer.cpp:69] Initializing prefetch
I0223 14:55:42.935431 6406 base_data_layer.cpp:72] Prefetch initialized.
I0223 14:55:42.935503 6406 net.cpp:150] Setting up mnist
I0223 14:55:42.935598 6406 net.cpp:157] Top shape: 100 1 28 28 (78400)
I0223 14:55:42.935681 6406 net.cpp:157] Top shape: 100 (100)
I0223 14:55:42.935739 6406 net.cpp:165] Memory required for data: 314000
I0223 14:55:42.935811 6406 layer_factory.hpp:77] Creating layer label_mnist_1_split
I0223 14:55:42.935909 6406 net.cpp:106] Creating Layer label_mnist_1_split
I0223 14:55:42.935940 6406 net.cpp:454] label_mnist_1_split <- label
I0223 14:55:42.935992 6406 net.cpp:411] label_mnist_1_split -> label_mnist_1_split_0
I0223 14:55:42.936048 6406 net.cpp:411] label_mnist_1_split -> label_mnist_1_split_1
I0223 14:55:42.936138 6406 net.cpp:150] Setting up label_mnist_1_split
I0223 14:55:42.936182 6406 net.cpp:157] Top shape: 100 (100)
I0223 14:55:42.936202 6406 net.cpp:157] Top shape: 100 (100)
I0223 14:55:42.936223 6406 net.cpp:165] Memory required for data: 314800
I0223 14:55:42.936246 6406 layer_factory.hpp:77] Creating layer conv1
I0223 14:55:42.936327 6406 net.cpp:106] Creating Layer conv1
I0223 14:55:42.936350 6406 net.cpp:454] conv1 <- data
I0223 14:55:42.936396 6406 net.cpp:411] conv1 -> conv1
I0223 14:55:42.936653 6406 net.cpp:150] Setting up conv1
I0223 14:55:42.936681 6406 net.cpp:157] Top shape: 100 20 24 24 (1152000)
I0223 14:55:42.936693 6406 net.cpp:165] Memory required for data: 4922800
I0223 14:55:42.936758 6406 layer_factory.hpp:77] Creating layer pool1
I0223 14:55:42.936802 6406 net.cpp:106] Creating Layer pool1
I0223 14:55:42.936823 6406 net.cpp:454] pool1 <- conv1
I0223 14:55:42.936872 6406 net.cpp:411] pool1 -> pool1
I0223 14:55:42.936938 6406 net.cpp:150] Setting up pool1
I0223 14:55:42.936962 6406 net.cpp:157] Top shape: 100 20 12 12 (288000)
I0223 14:55:42.936975 6406 net.cpp:165] Memory required for data: 6074800
I0223 14:55:42.936990 6406 layer_factory.hpp:77] Creating layer conv2
I0223 14:55:42.937044 6406 net.cpp:106] Creating Layer conv2
I0223 14:55:42.937064 6406 net.cpp:454] conv2 <- pool1
I0223 14:55:42.937110 6406 net.cpp:411] conv2 -> conv2
I0223 14:55:42.940639 6406 net.cpp:150] Setting up conv2
I0223 14:55:42.940696 6406 net.cpp:157] Top shape: 100 50 8 8 (320000)
I0223 14:55:42.940711 6406 net.cpp:165] Memory required for data: 7354800
I0223 14:55:42.940803 6406 layer_factory.hpp:77] Creating layer pool2
I0223 14:55:42.940876 6406 net.cpp:106] Creating Layer pool2
I0223 14:55:42.940910 6406 net.cpp:454] pool2 <- conv2
I0223 14:55:42.940970 6406 net.cpp:411] pool2 -> pool2
I0223 14:55:42.941061 6406 net.cpp:150] Setting up pool2
I0223 14:55:42.941093 6406 net.cpp:157] Top shape: 100 50 4 4 (80000)
I0223 14:55:42.941108 6406 net.cpp:165] Memory required for data: 7674800
I0223 14:55:42.941131 6406 layer_factory.hpp:77] Creating layer ip1
I0223 14:55:42.941184 6406 net.cpp:106] Creating Layer ip1
I0223 14:55:42.941210 6406 net.cpp:454] ip1 <- pool2
I0223 14:55:42.941262 6406 net.cpp:411] ip1 -> ip1
……
//输出准确率accuracy和损失loss
I0223 14:55:59.623518 6406 solver.cpp:408] Test net output #0: accuracy = 0.1309
I0223 14:55:59.623611 6406 solver.cpp:408] Test net output #1: loss = 2.31399 (* 1 = 2.31399 loss)
I0223 14:55:59.623705 6406 base_data_layer.cpp:115] Prefetch copied
I0223 14:55:59.625409 6410 data_layer.cpp:102] Prefetch batch: 1 ms.
I0223 14:55:59.625458 6410 data_layer.cpp:103] Read time: 0.188 ms.
I0223 14:55:59.625468 6410 data_layer.cpp:104] Transform time: 1.022 ms.
I0223 14:55:59.874915 6406 solver.cpp:229] Iteration 0, loss = 2.30499
I0223 14:55:59.874987 6406 solver.cpp:245] Train net output #0: loss = 2.30499 (* 1 = 2.30499 loss)
I0223 14:55:59.875026 6406 sgd_solver.cpp:106] Iteration 0, lr = 0.01
I0223 14:55:59.879196 6406 base_data_layer.cpp:115] Prefetch copied
I0223 14:55:59.881901 6410 data_layer.cpp:102] Prefetch batch: 2 ms.
I0223 14:55:59.881934 6410 data_layer.cpp:103] Read time: 0.413 ms.
I0223 14:55:59.881945 6410 data_layer.cpp:104] Transform time: 1.312 ms.
I0223 14:56:00.149636 6406 base_data_layer.cpp:115] Prefetch copied
I0223 14:56:00.151485 6410 data_layer.cpp:102] Prefetch batch: 1 ms.
I0223 14:56:00.151515 6410 data_layer.cpp:103] Read time: 0.195 ms.
I0223 14:56:00.151525 6410 data_layer.cpp:104] Transform time: 1.073 ms.
I0223 14:56:00.400404 6406 solver.cpp:458] Snapshotting to binary proto file examples/mnist/lenet_iter_3.caffemodel//保存当前训练模型快照
I0223 14:56:00.400461 6406 net.cpp:918] Serializing 9 layers
//保存当前solver的快照
I0223 14:56:00.434458 6406 sgd_solver.cpp:273] Snapshotting solver state to binary proto file examples/mnist/lenet_iter_3.solverstate
I0223 14:56:00.455166 6406 solver.cpp:323] Optimization Done.
I0223 14:56:00.455202 6406 caffe.cpp:222] Optimization Done.//程序运行结束
内容来自用户分享和网络整理，不保证内容的准确性，如有侵权内容，可联系管理员处理 
相关文章推荐
- 苹果手机的html 手势，点击事件
- JS之数组相关
- JS之函数相关
- JS之运算符相关
- JS之错误处理
- Bootstrap一些例子使用，持续更新...
- JS之程序结构
- JS之变量与常量
- JS之数据类型相关
- 韩顺平 javascript教学视频_学习笔记34_js正则表达式详解
- JS之概述
- javascript瀑布流式图片懒加载实例解析与优化
- 【CSS3】only-of-type选择器
- file_operations、inode、file
- 【CSS3】only-child选择器
- css 学习1 css基础
- javascript瀑布流式图片懒加载实例
- highcharts动态加载json数据
- HTML】Table边框使用总结 ，只显示你要显示的边框(内边框，外边框)
- 【CSS3】nth-last-of-type(n)选择器