YOLO-v4训练自己的数据集

论文传送门：https://arxiv.org/abs/2004.10934
参考文献： YoLov3训练自己的数据集（小白手册）
源项目README

github代码传送门:https://github.com/AlexeyAB/darknet
在这里插入图片描述

如何在MS COCOs数据集上评估YOLOv4的AP？

从MS COCO服务器下载并解压test-dev2017数据集:http://images.cocodataset.org/zips/test2017.zip
下载用于检测的图像列表，并将路径替换为您的: https://raw.githubusercontent.com/AlexeyAB/darknet/master/scripts/testdev2017.txt
下载yolov4.weights文件:https://drive.google.com/open?id=1cewMfusmPjYWbrnuJRuKhPMwRe_b9PaT

提供CSDN下载通道：https://download.csdn.net/download/qq_44166805/12378170

网盘下载通道：链接：https://pan.baidu.com/s/17XA84GBmSV7Thib1KCvQcA
提取码：l656

有积分的朋友可以选择CSDN来资助一下~
文件cfg/coco.data 应为：

classes= 80
train  = <replace with your path>/trainvalno5k.txt
valid = <replace with your path>/testdev2017.txt
names = data/coco.names
backup = backup
eval=coco

在可执行文件darknet所在文件夹里创建result文件夹

运行

1	./darknet detector valid cfg/coco.data cfg/yolov4.cfg yolov4.weights

将文件重命名为’ /results/coco_results.json”到“detections_test-dev2017_yolov4_results。将其压缩为:detections_test-dev2017_yolov4_results.zip

训练好的模型

yolov4.cfg - 245 MB: yolov4.weights paper Yolo v4
- width=608 height=608 in cfg: 65.7% mAP@0.5 (43.5% AP@0.5:0.95) - 34(R) FPS / 62(V) FPS - 128.5 BFlops
- width=512 height=512 in cfg: 64.9% mAP@0.5 (43.0% AP@0.5:0.95) - 45(R) FPS / 83(V) FPS - 91.1 BFlops
- width=416 height=416 in cfg: 62.8% mAP@0.5 (41.2% AP@0.5:0.95) - 55(R) FPS / 96(V) FPS - 60.1 BFlops
- width=320 height=320 in cfg: 60% mAP@0.5 ( 38% AP@0.5:0.95) - 63(R) FPS / 123(V) FPS - 35.5 BFlops

AlexeyAB改进项

提供window支持
相较于原版pjreddie版本darknet提升了训练速度
添加了二值化网络，XNOR(bit) ，速度快，准确率稍低https://github.com/AlexeyAB/darknet/blob/master/cfg/yolov3-tiny_xnor.cfg
提升7%通过将卷积层和BN层合并一个。
多GPU训练提升
修补了[reorg]层
添加了mAP, IOU，Precision-Recall计算

darknet detector map...
可以在训练过程中画loss图像
添加了根据自己数据集的anchor生成
提升视频检测，网络摄像头，opencv相关问题
提出了一个INT8的网络，提升了检测速度，但是准确率稍有下降https://github.com/AlexeyAB/yolo2_light
还有很多可以去看项目中的readme哦～

如何配置darknet

安装opencv以及NVIDIA显卡驱动and对应版本的CUDA（网上教程很多，不再赘述~）

下载安装darknet

1 2	git clone https://github.com/AlexeyAB/darknet cd darknet

打开Makefile

GPU=1 #需要GPU加速设为1,
CUDNN=0
CUDNN_HALF=0
OPENCV=1  #设为1
AVX=0
OPENMP=0
LIBSO=0
ZED_CAMERA=0 # ZED SDK 3.0 and above
ZED_CAMERA_v2_8=0 # ZED SDK 2.X

# 设置GPU=1 and CUDNN=1 开启GPU加速
# set CUDNN_HALF=1 to further speedup 3 x times (Mixed-precision on Tensor Cores) GPU: Volta, Xavier, Turing and higher
# 设置AVX=1和OPENMP=1 实现CPU加速 (如果报错设置 AVX=0)

设置完成后重新编译

1 2	make clean make -j8 #根据自己电脑决定线程，量力而行

测试安装是否成功（需要下载yolov4.weights）

1	./darknet detector test ./cfg/coco.data ./cfg/yolov4.cfg ./yolov4.weights data/dog.jpg -i 0 -thresh 0.25

在这里插入图片描述

安装成功

如何训练yolov4(训练你的自定义数据集):

创建 yolo-obj.cfg 文件，仿照yolov4-custom.cfg的样子 (也可以直接复制一份) ，并执行如下操作

（记得去掉注释）
- 把第三行batch改为batch=64
- 把subdivisions那一行改为 subdivisions=16
- 将max_batch更改为(数据集标签种类数（classes）*2000 但不小于4000 )
- 将第20的steps改为max_batch的0.8倍和0.9倍
- 把位于8-9行设为width=416 height=416 或者其他32的倍数:
- 将classes=80 改为你的类别数（有三个地方，969行，1056行，1143行）
- 改正[filters=255] 为 filters=(classes + 5)x3 （位置为查找yolo，每个yolo前的[convolutional]里，注意只修改最接近yolo的那个filters需要修改，一共应该有三处）
  
  贴出来修改后的cfg文件，仔细看注释哦

[net]
# Testing
batch=64  #显存不够的图像可以改为16,32等
subdivisions=16 #这里需要改
# Training
batch=64 # 显存不够的图像可以改为16,32等
subdivisions=16 # 这里需要改
width=416 #可以改为32的倍数
height=416 	#可以改为32的倍数
channels=3
momentum=0.949
decay=0.0005
angle=0
saturation = 1.5
exposure = 1.5
hue=.1

learning_rate=0.001
burn_in=1000
max_batches = 4000 #改为classses*2000，但不小于4000
policy=steps
steps=3200,3600	#改为max_batch的0.8倍和0.9倍
scales=.1,.1

#cutmix=1
mosaic=1

#:104x104 54:52x52 85:26x26 104:13x13 for 416

[convolutional]
batch_normalize=1
filters=32
size=3
stride=1
pad=1
activation=mish

# Downsample

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

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

[route]
layers = -2

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

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

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

[shortcut]
from=-3
activation=linear

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

[route]
layers = -1,-7

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

# Downsample

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

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

[route]
layers = -2

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

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

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

[shortcut]
from=-3
activation=linear

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

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

[shortcut]
from=-3
activation=linear

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

[route]
layers = -1,-10

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

# Downsample

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

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

[route]
layers = -2

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

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

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

[shortcut]
from=-3
activation=linear

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

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

[shortcut]
from=-3
activation=linear

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

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

[shortcut]
from=-3
activation=linear

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

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

[shortcut]
from=-3
activation=linear


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

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

[shortcut]
from=-3
activation=linear

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

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

[shortcut]
from=-3
activation=linear

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

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

[shortcut]
from=-3
activation=linear

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

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

[shortcut]
from=-3
activation=linear

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

[route]
layers = -1,-28

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

# Downsample

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

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

[route]
layers = -2

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

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

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

[shortcut]
from=-3
activation=linear


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

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

[shortcut]
from=-3
activation=linear


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

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

[shortcut]
from=-3
activation=linear


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

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

[shortcut]
from=-3
activation=linear


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

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

[shortcut]
from=-3
activation=linear


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

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

[shortcut]
from=-3
activation=linear


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

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

[shortcut]
from=-3
activation=linear

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

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

[shortcut]
from=-3
activation=linear

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

[route]
layers = -1,-28

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

# Downsample

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

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

[route]
layers = -2

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

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

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

[shortcut]
from=-3
activation=linear

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

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

[shortcut]
from=-3
activation=linear

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

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

[shortcut]
from=-3
activation=linear

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

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

[shortcut]
from=-3
activation=linear

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

[route]
layers = -1,-16

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

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

[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

### SPP ###
[maxpool]
stride=1
size=5

[route]
layers=-2

[maxpool]
stride=1
size=9

[route]
layers=-4

[maxpool]
stride=1
size=13

[route]
layers=-1,-3,-5,-6
### End SPP ###

[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
filters=256
size=1
stride=1
pad=1
activation=leaky

[upsample]
stride=2

[route]
layers = 85

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

[route]
layers = -1, -3

[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
filters=128
size=1
stride=1
pad=1
activation=leaky

[upsample]
stride=2

[route]
layers = 54

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

[route]
layers = -1, -3

[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=21  #这里改为 filters=(classes + 5)x3 
activation=linear


[yolo]
mask = 0,1,2
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
classes=2 #这里需要改成自己的classes
num=9
jitter=.3
ignore_thresh = .7
truth_thresh = 1
scale_x_y = 1.2
iou_thresh=0.213
cls_normalizer=1.0
iou_normalizer=0.07
iou_loss=ciou
nms_kind=greedynms
beta_nms=0.6


[route]
layers = -4

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

[route]
layers = -1, -16

[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=21  #这里改为 filters=(classes + 5)x3 
activation=linear


[yolo]
mask = 3,4,5
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
classes=2   #这里需要改成自己的classes数目
num=9
jitter=.3
ignore_thresh = .7
truth_thresh = 1
scale_x_y = 1.1
iou_thresh=0.213
cls_normalizer=1.0
iou_normalizer=0.07
iou_loss=ciou
nms_kind=greedynms
beta_nms=0.6


[route]
layers = -4

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

[route]
layers = -1, -37

[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
filters=21 #这里改为 filters=(classes + 5)x3 
activation=linear


[yolo]
mask = 6,7,8
anchors = 12, 16, 19, 36, 40, 28, 36, 75, 76, 55, 72, 146, 142, 110, 192, 243, 459, 401
classes=2   #这里需要改成自己的classes数目
num=9
jitter=.3
ignore_thresh = .7
truth_thresh = 1
random=1
scale_x_y = 1.05
iou_thresh=0.213
cls_normalizer=1.0
iou_normalizer=0.07
iou_loss=ciou
nms_kind=greedynms
beta_nms=0.6

处理数据集,

在scripts文件夹下创建文件夹VOCdevkit（因为scripts文件夹下有vov_label.py文件，它的作用下面会说，下面创建的文件也跟它有关），根据下图在VOCdevkit文件夹下创建文件，并放入相应的数据

VOCdevkit 
——VOC2020        #文件夹的年份可以自己取,但是要与你其他文件年份一致，看下一步就明白了
————Annotations  #放入所有的xml文件
————ImageSets    
——————Main       #放入train.txt,val.txt文件
————JPEGImages   #放入所有的图片文件

#Main中的文件分别表示test.txt是测试集，train.txt是训练集，val.txt是验证集，trainval.txt是训练和验证集，这里我只用了train和valid

在这里插入图片描述

其中Main中的txt文件是要写文件名，比如train.txt里写的是用来训练的图片的文件名(不包含后缀，只是文件名哦！！！)，这个文件可以找代码生成（下面的python代码可以用），代码的话看懂他的作用，特别的文件的路径之类的，根据自己的情况修改下，就可以用:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
import os
from os import listdir, getcwd
from os.path import join
if __name__=='__main__':
    source_folder='/home/linux/darknet-master/scripts/VOCdevkit/VOC2020/JPEGImages'
    dest='/home/linux/darknet-master/scripts/VOCdevkit/VOC2020/ImageSets/Main/train.txt'   # train.txt文件路径
    dest2 = '/home/linux/darknet-master/scripts/VOCdevkit/VOC2020/ImageSets/Main/val.txt'  # val.txt文件路径
    file_list=os.listdir(source_folder)
    train_file=open(dest,'a')
    val_file = open(dest2, 'a')
    file_num=0
    for file_obj in file_list:
        file_path=os.path.join(source_folder,file_obj)
        file_name,file_extend=os.path.splitext(file_obj)
        file_num=file_num+1
        
        if(file_num%50==0): #每隔50张选取一张验证集
            val_file.write(file_name+'\n')
        else:
            train_file.write(file_name+'\n')
train_file.close()
val_file.close()

修改voc_label.py，这个文件就是根据Main中txt里的文件名，生成相应的txt，里面存放的是它们的路径

sets=[('2020', 'train'), ('2020', 'val')]   #这里要与Main中的txt文件一致
 
classes = ["hat", "person"]        #你所标注的表签名，第一步中已经说过
 
os.system("cat 2020_train.txt 2020_val.txt> train.txt")  #文件最后一句，意思是要将生成的文件合并，所以很据需要修改，这里的年份都是一致的，简单理解下代码应该会懂，不要盲目修改

1	python voc_label.py #保存后运行

运行后会生成2020_train.txt、2020_val.txt、train.txt

复制data目录下的voc.name，改为*.name（例如safe.name),里面写标签的名字，例如
1
2
hat
person

复制cfg文件夹下的voc.data，重命名为*.data（例如：safe.data):

classes= 2 							#classes为训练样本集的类别总数
train  = scripts/train.txt 			#train的路径为训练样本集所在的路径，前面生成的
valid  = scripts/test.txt			#valid的路径为验证样本集所在的路径，前面生成的
names = data/safe.names				#names的路径为***.names文件所在的路径 
backup = backup/					#模型保存地点

下载预训练权重

for yolov4.cfg, yolov4-custom.cfg (162 MB):yolov4.conv.137

为方便国内读者下载，提供百度网盘：链接：https://pan.baidu.com/s/1rAzuhN6-mZwPOLyxF-7g3g
提取码：8b16

有条件的可以用积分下载一下～（没积分的孩子在线卑微）

链接：https://download.csdn.net/download/qq_44166805/12378178

开始训练

1	./darknet detector train data/obj.data yolo-obj.cfg yolov4.conv.137 -map

-map是输出MAP值，可以不加

断点继续训练

在每100次迭代之后，您可以停止，然后从这一点开始培训。例如，在2000次迭代之后，您可以停止训练，然后使用以下命令开始继续训练:
1
./darknet detector train data/obj.data yolo-obj.cfg backup/yolo-obj_2000.weight
每1000次迭代保存一次，并自动保存当前最优模型。

模型检验

例如：检测mAP值

1	./darknet detector map data/obj.data yolo-obj.cfg backup/yolo-obj_7000.weights

用模型检测图片：

1	./darknet detector test cfg/obj.data cfg/obj.cfg backup/obj.weights test.jpg -i 0 -thresh 0.25 -gpus 0,1,2,3

用模型检测视频流：

1	./darknet detector demo cfg/coco.data cfg/yolov4.cfg yolov4.weights test.mp4 -dont_show -ext_output -gpus 0,1,2,3

服务器玩家可以在上述指令后加入

1	-dont_show -mjpeg_port 9090 #9090可以自己随意指定另一个

然后利用端口监听就可以看到了（具体操作各个软件不一样），这里我用的xshell6的配置如下：
在这里插入图片描述
然后打开浏览器，输入127.0.0.1:9090就可以啦。
（Ps：训练时用nohup会很舒服）

如何多GPU训练？

先在1个GPU上进行1000次迭代训练:

1	./darknet detector train cfg/coco.data cfg/yolov4.cfg yolov4.conv.137

然后停止并使用刚刚训练的训练的模型/backup/yolov4_1000.weights 进行多GPU训练（这里用了四块）：
1
./darknet detector train cfg/coco.data cfg/yolov4.cfg /backup/yolov4_1000.weights -gpus 0,1,2,3
对于小数据集降低学习率会更好，为4个gpu设置’learning_rate = 0.00025 (即learning_rate = 0.001 / gpu)。在这种情况下，还要在cfg文件中增加4倍的’burn_in = 和 max_batch = 。例如，使用’ burn_in = 4000 而不是’1000 。如果设置了’policy=steps ，那么steps= 也一样。

如何批量测试？

制作test.txt文件:
制作一个带绝对路径的批量测试文档，例如如下所示：
在这里插入图片描述
话不多说，上代码：

import os
from os import listdir, getcwd
from os.path import join
if __name__=='__main__':
    source_folder='/home/linux/darknet-master/testdata/JPEGImages/' #测试集所在位置
    dest = '/home/linux/darknet-master/data/test.txt'  # test.txt文档要保存的位置
    file_list=os.listdir(source_folder)
    test_file= open(dest,'a') #追加写打开
    file_num=0
    for file_obj in file_list:
        file_path=os.path.join(source_folder,file_obj)
        file_name,file_extend=os.path.splitext(file_obj)
        file_num=file_num+1
        test_file.write(source_folder+file_name+'.jpg'+'\n')
test_file.close()

运行darknet进行批量测试
如果你只是想批量测试data/test.txt并把结果保存至result.txt，可以用下面这个指令，当然，如果想看图片识别效果可以将-dont_show参数去掉

1	./darknet detector test cfg/coco.data cfg/yolov4.cfg yolov4.weights -dont_show -ext_output < data/test.txt > result.txt

darknet还提供了一种“假标定”方法来扩充数据集,处理图像列表data/new_train.txt ，并将每个图像的检测结果以Yolo训练标注格式保存为到.txt (可以用来增加训练数据) :

1	./darknet detector test cfg/coco.data cfg/yolov4.cfg yolov4.weights -thresh 0.25 -dont_show -save_labels < data/new_train.txt

注：这种方法生成的txt可在你的训练集txt储存的地方找到