yolov3.cfg文件详细说明

     本文对darknet源码中的yolov-voc.cfg重要参数详细解读说明，同时参考多个博客并融合个人理解，供初学者学习参考，谢谢！

[net]     #net 为特殊的层，配置整个网络
#Testing
#batch=1
#subdivisions=1
#Training
#batch=64,sub=16表示训练过程中一次性加载64张图片进内存，然后分16次完成前向传播，每次4张，
#前向传播的循环过程中累加loss求平均，待64张图片都完成前向传播后，再反向传播更新参数
#batch越大，训练效果越好，subdivision越大，占用内存压力越小
#测试时batch=1,sub=1
batch=64   
subdivisions=16
width=416 #输入网络的图片宽，值越大，小目标识别效果越好，只可设置 为32的倍数
height=416 #输入网络的图片高，值越大，小目标识别效果越好，只可设置为32的倍数
channels=3 #输入图片的通道数
momentum=0.9 #最优化方法中的动量参数，影响梯度下降到最优的速度
decay=0.0005 #权重衰减正则项，防止过拟合
angle=0 #数据增强参数，通过旋转角度来生成更多训练样本
saturation = 1.5 #数据增强参数，通过调整饱和度来生成更多训练样本
exposure = 1.5   #数据增强参数，通过调整曝光量来生成更多训练
hue=.1  #数据增强参数，通过调整色调来生成更多训练
#学习率决定着权值更新的速度，设置得太大会使结果超过最优值，太小会使下降速度过慢。
#手动调整时，刚开始可将学习率设置高一些，学习率以 0.01 ~ 0.001为宜。一定轮数过后：逐渐减缓，
#可设置动态变化的学习率。接近训练结束：学习速率的衰减应该在100倍以上
#若loss波动过大则学习率过大,可适当减小(1/5,1/10);若几乎不变，说明已收敛或陷入局部最优，可增大。
learning_rate=0.001 
burn_in=1000 #迭代次数大于burn_in时，学习率采用policy更新方式
max_batches = 50200 #训练次数达到max_batches后停止训练
policy=steps #学习率策略的调整，constant,steps,exp,ploy,step,sig,RANDOM,constant
#step和scale时设置学习率的变化，迭代到40000次时，学习率衰减10倍，迭代到40000次时，学习率再衰减10倍
steps=40000,45000
scales=.1,.1

[convolutional]#卷积层配置说明
batch_normalize=1 #是否进行BN处理
filters=32 #卷积核个数，也是输出通道数
size=3 #卷积核尺寸
stride=1 #卷积步长
pad=1 #是否进行0 padding，pad为1，则padding大小为卷积核size/2 向下取整
activation=leaky #激活函数类型

# Downsample

[convolutional]
batch_normalize=1
filters=64
size=3
stride=2
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=32
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=64
size=3
stride=1
pad=1
activation=leaky

[shortcut] #shortcut层配置说明
from=-3    #与向前多少层进行融合，-3表示向前第三层
activation=linear #层激活函数

# Downsample

[convolutional]
batch_normalize=1
filters=128
size=3
stride=2
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=64
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

[convolutional]
batch_normalize=1
filters=64
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

# Downsample

[convolutional]
batch_normalize=1
filters=256
size=3
stride=2
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

# Downsample

[convolutional]
batch_normalize=1
filters=512
size=3
stride=2
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

# Downsample

[convolutional]
batch_normalize=1
filters=1024
size=3
stride=2
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
pad=1
activation=leaky

[shortcut]
from=-3
activation=linear

######################

[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=1024
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=1024
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=1024
activation=leaky

[convolutional]
size=1
stride=1
pad=1
#与yolo层classes相关联 filters=num*(classes+5)，5表示框坐标x,y,w,h和置信度c,num表示yolo中
#每个cell预测的框的个数，yolov3中为3，classes为需要识别的类别数
filters=75 
activation=linear

[yolo]
mask = 6,7,8 #不同尺度上对应的anchor box 索引
#anchors 大小
anchors = 10,13,  16,30,  33,23,  30,61,  62,45,  59,119,  116,90,  156,198,  373,326
classes=20 #目标类别数目
num=9 #每个gridcell 总共预测几个box,和anchors数量一致
jitter=.3 #数据增强手段，此处jitter为随机调整宽高比的范围
#参与计算的IOU阈值大小，当预测的检测框与ground truth的IOU大于阈值时，参与loss计算，否则不参与
ignore_thresh = .5
truth_thresh = 1
#random表示训练过程中是否调整输入图像的尺寸，random=0表示图像尺寸不变，如果显存小，可设置为0
#random=1表示训练每迭代10次，随机调整输入图像的尺寸，每次迭代图片大小随机从320到608，步长为32
random=1
#路由层可设置为1个或2个值，是引出作用，不涉及卷积运算
#layers一个值时,输出改值索引的图层的特征图，如-4表示route层输出倒数第四层的特征图
#当layers两个值时，返回由其值所用的层的拼接特征图，如-1，-10表示route输出前1和前第10层沿深度维度拼接后图
[route]
layers = -4

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[upsample]
stride=2

[route]
layers = -1, 61

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=512
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=512
activation=leaky

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=512
activation=leaky

[convolutional]
size=1
stride=1
pad=1
filters=75
activation=linear

[yolo]
mask = 3,4,5
anchors = 10,13,  16,30,  33,23,  30,61,  62,45,  59,119,  116,90,  156,198,  373,326
classes=20
num=9
jitter=.3
ignore_thresh = .5
truth_thresh = 1
random=1

[route]
layers = -4

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky

[upsample]
stride=2

[route]
layers = -1, 36

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=256
activation=leaky

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=256
activation=leaky

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=256
activation=leaky

[convolutional]
size=1
stride=1
pad=1
filters=75
activation=linear

[yolo]
mask = 0,1,2
anchors = 10,13,  16,30,  33,23,  30,61,  62,45,  59,119,  116,90,  156,198,  373,326
classes=20
num=9
jitter=.3
ignore_thresh = .5
truth_thresh = 1
random=1