使用darknet框架的imagenet数据分类预训练操作

砸漏

发布于 2020-10-21 15:19:17

9250

发布于 2020-10-21 15:19:17

文章被收录于专栏：恩蓝脚本

最近一段时间一直在研究yolo物体检测，基于网络上很少有yolo的分类预训练和yolo9000的联合数据的训练方法，经过本人的真实实验，对这两个部分做一个整理（本篇介绍yolo的分类预训练）

1、数据准备

1000类的Imagenet图片数据

因为Imagenet不同的类别数据都是单独放在一个文件夹中，并且有特定的命名，如‘n00020287’，所以在做分类时我们不需要去制作特定的标签，只要训练的图片的path中包含自身的类别标签，而不含有其他类的标签即可。

制作用于训练的数据列表*classf_list.txt

2、分类标签制作

制作所有类别的标签列表new_label.txt和标签对应的类别名称的列表new_name.txt

new_label.txt

new_name.txt（训练时不需要，但是测试时可以显示出具体的类别）

3、修改cfg/.data配置文件（*classf.data）

classes=1000
train =/home/research/disk2/wangshun/yolo1700/darknet/coco/filelist/classf_list.txt
labels=data/new_label.txt
names=data/new_name.txt
backup=backup
top=5

修改网络配置文件(classf.cfg)

[net]
#Training
batch=64
subdivisions=1
width=416
height=416
channels=3
momentum=0.9
decay=0.0005
angle=0
saturation = 1.5
exposure = 1.5
hue=.1
max_crop = 512
learning_rate=0.001
burn_in=1000
max_batches = 1000000000
policy=steps
steps=350000,500000,750000,1200000
scales=.1,.1,.1,.1
[convolutional]
batch_normalize=1
filters=16
size=3
stride=1
pad=1
activation=leaky
[maxpool]
size=2
stride=2
[convolutional]
batch_normalize=1
filters=32
size=3
stride=1
pad=1
activation=leaky
[maxpool]
size=2
stride=2
[convolutional]
batch_normalize=1
filters=64
size=3
stride=1
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
[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=64
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
[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=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
[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
#######
[convolutional]
batch_normalize=1
size=1
stride=1
pad=1
filters=128
activation=leaky
[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=256
activation=leaky
[convolutional]
batch_normalize=1
size=3
stride=1
pad=1
filters=256
activation=leaky
[convolutional]
filters=1000
size=1
stride=1
pad=1
activation=leaky
[avgpool]
[softmax]
groups = 1
[cost]
type=sse

当然中间的网络层是我自己修改的网络。

5. 训练