用自己的数据训练yolov3

Train YOLO V3 with my own data

Operating System: Linux on Ubuntu with CUDA 8.0 & OPENCV 3.2.0

 

Since YOLO V3 has a better performance on small objection detection, and we add a new class—wrist, we switched to YOLO V3. There is a great tutorial to guide me, and below is what I’ve done to train my own data.

 

Step 1. Prepare data

 

Label your pictures, I used labelImg to get this done! I labeled four classes objection: hand, skull, elbow joint, wrist; 3631 pictures in total.

Guided by this blog, txt files for training and validation can be get by the create_list.py and voc_label.py. There are to things you should remember: (1) change the classes in line 7 in voc_label.py. (2) change the comment from line 47 to 50 to get val.txt like below.

#image_ids = open('ImageSets/Main/train.txt').read().strip().split()  
#list_file = open('train.txt', 'w')  
image_ids = open('ImageSets/Main/val.txt').read().strip().split()  
list_file = open('val.txt', 'w') 

 If you did everything right, you’ll get a folder like this:

 

 

Step 2.  Change the configuration files

 

Three config files should be changed: cfg/voc.data, cfg/yolo-voc.cfg.

 

cfg/yolov3-voc.data

According to the directory of your file, rewrite the voc.data file. In my project, I only labeled one feature: hand, so I changed the classes number from 20 to 1. I saved all files related to this project under a folder named mytrain, and got my train.txt and val.txt files guided by this blog, so I changed my voc.data to this:

classes= 4  
train = /home/tec/darknet/mytrain/train.txt  
valid = /home/tec/darknet/mytrain/val.txt  
names = data/voc.names  
backup = /home/tec/darknet/backup  

Don’t forget to create the voc.names file under the folder named data, in hand.names text:

hand   
skull  
elbow joint   
wrist  

 

cfg/yolo-voc.cfg

 

you can change it according to the commands in below.

 

Notes:

组合键ctrl+f找yolo，可以看到3处需要修改的filters

filters = (classes+ 5)* 3, in this case filters=(4+5)*3=27

If the video memory is very small, set random = 0 to turn off multiscale training, else set random=1.  

[net]  
# Testing  
# batch=1  
# subdivisions=1  
# Training  
batch=64  
subdivisions=8  
  
......  
  
[convolutional]  
size=1  
stride=1  
pad=1  
filters=27 #filters = (classes+ 5)* 3  
activation=linear  
  
[yolo]  
mask = 6,7,8  
anchors = 10,13,  16,30,  33,23,  30,61,  62,45,  59,119,  116,90,  156,198,  373,326  
classes=4  
num=9  
jitter=.3  
ignore_thresh = .5  
truth_thresh = 1  
random=0 #If the video memory is very small, set random = 0 ato turn off multiscale training, else set random=1.  
......  
  
[convolutional]  
size=1  
stride=1  
pad=1  
filters=27 #filters = (classes+ 5)* 3  
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=4###20  
num=9  
jitter=.3  
ignore_thresh = .5  
truth_thresh = 1  
random=0###1  
......  
  
[convolutional]  
size=1  
stride=1  
pad=1  
filters=27 #filters = (classes+ 5)* 3  
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=4  
num=9  
jitter=.3  
ignore_thresh = .5  
truth_thresh = 1  
random=0###1  

Step 4 Download the pre-trained weight

wget https://pjreddie.com/media/files/darknet53.conv.74  

Step 5. Train and test

 

 Train

./darknet detector train cfg/voc.data cfg/yolov3-voc.cfg darknet53.conv.74  

Test

./darknet detector test cfg/voc.data cfg/yolov3-voc.cfg backup/yolov3-voc.backup data/test/7.jpg #test picture 

 

./darknet detector demo cfg/voc.data cfg/yolov3-voc.cfg backup/yolov3-voc.backup  #test by video  

批量测试图片，参考博客1,博客2；

 

YOLOV4的编译和训练

https://zhuanlan.zhihu.com/p/340025123