版权声明:未经博主允许,不得转载! https://blog.csdn.net/u011681952/article/details/86157060
Convolution Layer是CNN中最常见最重要的特征提取层,形式多种多样
首先我们先看一下 InnerProductParameter
message ConvolutionParameter {
optional uint32 num_output = 1; // The number of outputs for the layer 输出特征图个数(常以特征图为单位)
optional bool bias_term = 2 [default = true]; // whether to have bias terms 是否使用bias项
// Pad, kernel size, and stride are all given as a single value for equal
// dimensions in all spatial dimensions, or once per spatial dimension.
repeated uint32 pad = 3; // The padding size; defaults to 0 填充大小(像素为单位)
repeated uint32 kernel_size = 4; // The kernel size 卷积核大小
repeated uint32 stride = 6; // The stride; defaults to 1 步幅大小
// Factor used to dilate the kernel, (implicitly) zero-filling the resulting
// holes. (Kernel dilation is sometimes referred to by its use in the
// algorithme à trous from Holschneider et al. 1987.)
repeated uint32 dilation = 18; // The dilation; defaults to 1 空洞大小 默认1(空洞卷积中常用,如某些语义分割网络)
// For 2D convolution only, the *_h and *_w versions may also be used to
// specify both spatial dimensions.
//2D卷积,当高宽不一致时,常常用下列参数
optional uint32 pad_h = 9 [default = 0]; // The padding height (2D only) 高度填充大小
optional uint32 pad_w = 10 [default = 0]; // The padding width (2D only) 宽度填充大小
optional uint32 kernel_h = 11; // The kernel height (2D only) 卷积核高度
optional uint32 kernel_w = 12; // The kernel width (2D only) 卷积核宽度
optional uint32 stride_h = 13; // The stride height (2D only) y轴步幅
optional uint32 stride_w = 14; // The stride width (2D only) x轴步幅
optional uint32 group = 5 [default = 1]; // The group size for group conv 组卷积 默认1 实例见shufflenet
optional FillerParameter weight_filler = 7; // The filler for the weight 卷积权重参数
optional FillerParameter bias_filler = 8; // The filler for the bias 偏置项参数
enum Engine {
DEFAULT = 0;
CAFFE = 1;
CUDNN = 2;
}
optional Engine engine = 15 [default = DEFAULT];
// The axis to interpret as "channels" when performing convolution.
// Preceding dimensions are treated as independent inputs;
// succeeding dimensions are treated as "spatial".
// With (N, C, H, W) inputs, and axis == 1 (the default), we perform
// N independent 2D convolutions, sliding C-channel (or (C/g)-channels, for
// groups g>1) filters across the spatial axes (H, W) of the input.
// With (N, C, D, H, W) inputs, and axis == 1, we perform
// N independent 3D convolutions, sliding (C/g)-channels
// filters across the spatial axes (D, H, W) of the input.
optional int32 axis = 16 [default = 1]; 卷积轴,默认通道(3D卷积中,有时序轴卷积情况)
// Whether to force use of the general ND convolution, even if a specific
// implementation for blobs of the appropriate number of spatial dimensions
// is available. (Currently, there is only a 2D-specific convolution
// implementation; for input blobs with num_axes != 2, this option is
// ignored and the ND implementation will be used.)
//是否强制使用一般的ND卷积,即使对于具有适当空间维数的blob有特定的实现。(目前只有2d特有的卷积实现;对于num_axes != 2的输入blob,将忽略此选项,并使用ND卷积)
optional bool force_nd_im2col = 17 [default = false];
}
卷积形式太多,一时难以收集全,先举个例子,以后慢慢更新
例如在MobileNet中:
layer {
name: "conv6_3/dwise"
type: "Convolution"
bottom: "conv6_3/expand/bn"
top: "conv6_3/dwise"
param {
lr_mult: 1
decay_mult: 1
}
convolution_param {
num_output: 960 //输出个数
bias_term: false //不使用bias项
pad: 1 //填充1个像素
kernel_size: 3 //卷积核大小3*3
group: 960 //组个数 则一组个数:num_output/ group (必须整除)
weight_filler {
type: "msra"
}
engine: CAFFE
}
}