darknet用公共数据集训练自己的模型

目录结构如下

.txt文件是由train_label.py文件生成的,其他的文件需要根据自己选定的版本yolo版本进行选择.(这里以yolov3-tiny为例)

### 修改配置train_label.py保修需要的类.然后执行

python train_label.py # 执行后即可出线`.txt`文件,如果遇到报错请查看数据集的位置是否正确.

### 修改v3_tiny_obj.names文件 ### 修改v3_tiny_obj.data文件 ### 生成yolov3-tiny.conv.15文件 (预训练权重的提取)

darknet partial data/voc/yolov3-tiny_obj.cfg data/voc/yolov3-tiny.weights data/voc/yolov3-tiny.conv.15 15

#### 修改yolov3-tiny_obj.cfg 锚点计算,普通版本9, tiny 6

darknet detector calc_anchors data/voc/v3_tiny_obj.data -num_of_clusters 6 -width 416 -height 416

修改内容如下

[net]
# Testing
#batch=1
#subdivisions=1
# Training
# 根据电脑性能修改大小,  batch = 8*n
batch=64
# 根据电脑性能调整大小
subdivisions=2    
width=416
height=416
channels=3
momentum=0.9
decay=0.0005
angle=0
saturation = 1.5
exposure = 1.5
hue=.1

learning_rate=0.001
burn_in=1000
#max_batches = 500200
#steps=400000,450000
# 训练步数
max_batches = 10000
policy=steps
# 训练到相应次数后学习率变化
steps=4000,8000
scales=.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

[maxpool]
size=2
stride=2

[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

[maxpool]
size=2
stride=2

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

[maxpool]
size=2
stride=1

[convolutional]
batch_normalize=1
filters=1024
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]
size=1
stride=1
pad=1
# classes:训练类别数:   filters=3*((classes+5)
filters=18
activation=linear



[yolo]
mask = 3,4,5
# 锚点位置计算
anchors = 23, 63,  57,138,  90,255, 160,179, 165,321, 310,361
# 根据自己设定的类别数修改
classes=1
num=6
jitter=.3
ignore_thresh = .7
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, 8

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

[convolutional]
size=1
stride=1
pad=1
# classes:训练类别数:   filters=3*((classes+5)
filters=18
activation=linear

[yolo]
# mask = 0,1,2
mask = 1,2,3
# 锚点位置计算
anchors = 23, 63,  57,138,  90,255, 160,179, 165,321, 310,361
classes=1
num=6
jitter=.3
ignore_thresh = .7
truth_thresh = 1
random=1

开始训练

cpu训练:

在darknet.exe所在的路径下打开cmd执行一下命令!!!!!!!

darknet_no_gpu detector train data/voc/v3_tiny_obj.data data/voc/yolov3-tiny_obj.cfg data/voc/yolov3-tiny.conv.15

在这里插入图片描述

gpu训练:

修改工程文件中的makeFile文件 #### makeFile中修改内容如下 ### 开始训练 ## 在darknet.exe所在的路径下打开cmd执行一下命令!!!!!!!

darknet detector train data/voc/v3_tiny_obj.data data/voc/yolov3-tiny_obj.cfg data/voc/yolov3-tiny.conv.15

训练效果: ### 训练完权重文件存放路径