yolo-voc.2.0.cfg 参数解析

一、参数解析

[net]
batch=64         # number of images pushed with a forward pass through the network
subdivisions=8   # 源码中的图片数量int imgs = net.batch * net.subdivisions * ngpus，按subdivisions大小分批进行训练 
height=416       # height of input image
width=416        # width of input image
channels=3       # channel of input image
momentum=0.9     # CNN-梯度下降法中一种常用的加速技术
decay=0.0005     # CNN-防止过拟合

# 对于每次迭代训练，YOLOv2会基于角度(angle)，饱和度(saturation)，曝光(exposure)，色调(hue)产生新的训练图片 
angle=0          # 图片角度变化，单位为度，假如angle=5，就是生成新图片的时候随机旋转-5~5度
saturation = 1.5 # 饱和度变化大小，1到1.5倍
exposure = 1.5   # 曝光变化大小，1到1.5倍
hue=.1           # 色调变化范围，-0.1到0.1 

learning_rate=0.0001  # 学习率 
max_batches = 45000   # 最大迭代次数 
policy=steps          # 调整学习率的policy：CONSTANT, STEP, EXP, POLY，STEPS, SIG, RANDOM
steps=100,25000,35000 # 根据batch_num调整学习率，若steps=100,25000,35000，则在迭代100次，25000次，35000次时学习率发生变化，该参数与policy中的steps对应
scales=10,.1,.1       # 相对于当前学习率的变化比率，累计相乘，与steps中的参数个数保持一致

[convolutional]
batch_normalize=1     # 是否做BN-batch_normalize
filters=32
size=3
stride=1
pad=1
activation=leaky
# 激活函数-activation 包括logistic，loggy，relu，elu，relie，plse，hardtan，lhtan，linear，ramp，leaky，tanh，stair.

[maxpool]
size=2
stride=2

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

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
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

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
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

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
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

[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

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=1024
size=3
stride=1
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

[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


#######

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

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

# the route layer is to bring finer grained features in from earlier in the network 
[route]
layers=-9

# the reorg layer is to make these features match the feature map size at the later layer;
# The end feature map is 13x13, the feature map from earlier is 26x26x512. 
# The reorg layer maps the 26x26x512 feature map onto a 13x13x2048 feature map so that it can be concate_nated with the feature maps at 13x13 resolution.
[reorg]
stride=2

[route]
layers=-1,-3

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

[convolutional]
size=1
stride=1
pad=1
filters=125
activation=linear

[region]
# anchors: 预测框的初始宽高，第一个是w，第二个是h，总数量是num*2.
# YOLOv2作者说anchors是使用K-MEANS获得，其实就是计算出哪种类型的框比较多，可以增加收敛速度，如果不设置anchors，默认是0.5.
anchors = 1.08,1.19,  3.42,4.41,  6.63,11.38,  9.42,5.11,  16.62,10.52
bias_match=1 # 如果为1，计算best iou时，预测宽高强制与anchors一致
classes=20   # 类别数量 
coords=4     # BoundingBox的tx,ty,tw,th，tx与ty是相对于左上角的grid，同时是当前grid的比例，tw与th是宽度与高度取对数
num=5        # 每个grid预测的BoundingBox个数
softmax=1    # 如果为1，使用softmax 
jitter=.2    # 利用数据抖动产生更多数据抑制过拟合.YOLOv2中使用的是crop，filp，以及net层的angle，flip是随机的，crop就是jitter的参数，tiny-yolo-voc.cfg中jitter=.2，就是在0~0.2中进行crop. 
rescore=1    # 决定使用哪种方式计算IOU的误差，为1时，使用当前best iou计算，为0时，使用1计算 

# *_scale是YOLOv1论文中cost function的权重，哪一个更大，每一次更新权重的时候，对应方面的权重更新相对比重更大
object_scale=5
noobject_scale=1
class_scale=1
coord_scale=1

absolute=1
thresh = .6 # 决定是否需要计算IOU误差的参数，大于thresh，IOU误差不会夹在cost function中 
random=0    # 如果为1每次迭代图片大小随机从320到608，步长为32，如果为0，每次训练大小与输入大小一致

二、训练log中各参数的意义

Region Avg IOU：平均的IOU，代表预测的bounding box和ground truth的交集与并集之比，期望该值趋近于1。 
Class:是标注物体的概率，期望该值趋近于1. 
Obj：期望该值趋近于1. 
No Obj:期望该值越来越小但不为零. 
Avg Recall：期望该值趋近1 
avg：平均损失，期望该值趋近于0

yolo-voc.2.0.cfg 参数解析

一、参数解析

二、训练log中各参数的意义

三、参考

猜你喜欢