pytorch—yolov3训练自己的数据集

注意：本篇博客直接使用VOC2007数据集

1.数据集

Labelimg软件构建数据集，Labelimg项目地址：https://github.com/tzutalin/labelImg，Labelimg快捷键：

Ctrl + u    Load all of the images from a directory
Ctrl + r    Change the default annotation target dir
Ctrl + s    Save
Ctrl + d    Copy the current label and rect box
Space    Flag the current image as verified
w    Create a rect box
d    Next image
a    Previous image
del    Delete the selected rect box
Ctrl++    Zoom in
Ctrl--    Zoom out
↑→↓←    Keyboard arrows to move selected rect box
voc2007数据集目录结构 ：

               ----voc2007

                           ----Annotations

                           ----ImageSets

                                        ----Main

                           ----JPEGImages

在根目录下新建makeTXT.py，将数据集划分，并且在Main文件夹下构建4个TXT：train.txt，test.txt，trainval.txt，val.txt。代码如下：

import os
import random

trainval_percent = 0.1
train_percent = 0.9
xmlfilepath = 'data/Annotations'
txtsavepath = 'data/ImageSets'
total_xml = os.listdir(xmlfilepath)

num = len(total_xml)
list = range(num)
tv = int(num * trainval_percent)
tr = int(tv * train_percent)
trainval = random.sample(list, tv)
train = random.sample(trainval, tr)

ftrainval = open('data/ImageSets/trainval.txt', 'w')
ftest = open('data/ImageSets/test.txt', 'w')
ftrain = open('data/ImageSets/train.txt', 'w')
fval = open('data/ImageSets/val.txt', 'w')

for i in list:
    name = total_xml[i][:-4] + '\n'
    if i in trainval:
        ftrainval.write(name)
        if i in train:
            ftest.write(name)
        else:
            fval.write(name)
    else:
        ftrain.write(name)

ftrainval.close()
ftrain.close()
fval.close()
ftest.close()

在 根目录下新建voclabel.py，生成labels。代码如下：

import xml.etree.ElementTree as ET
import pickle
import os
from os import listdir, getcwd
from os.path import join
sets = ['train', 'test','val']
classes = ["aeroplane", "bicycle", "bird", "boat", "bottle", "bus", "car", "cat", "chair", "cow", "diningtable", "dog", "horse", "motorbike", "person", "pottedplant", "sheep", "sofa", "train", "tvmonitor"]

def convert(size, box):
    dw = 1. / size[0]
    dh = 1. / size[1]
    x = (box[0] + box[1]) / 2.0
    y = (box[2] + box[3]) / 2.0
    w = box[1] - box[0]
    h = box[3] - box[2]
    x = x * dw
    w = w * dw
    y = y * dh
    h = h * dh
    return (x, y, w, h)
def convert_annotation(image_id):
    in_file = open('data/Annotations/%s.xml' % (image_id))
    out_file = open('data/labels/%s.txt' % (image_id), 'w')
    tree = ET.parse(in_file)
    root = tree.getroot()
    size = root.find('size')
    w = int(size.find('width').text)
    h = int(size.find('height').text)
    for obj in root.iter('object'):
        difficult = obj.find('difficult').text
        cls = obj.find('name').text
        if cls not in classes or int(difficult) == 1:
            continue
        cls_id = classes.index(cls)
        xmlbox = obj.find('bndbox')
        b = (float(xmlbox.find('xmin').text), float(xmlbox.find('xmax').text), float(xmlbox.find('ymin').text),
             float(xmlbox.find('ymax').text))
        bb = convert((w, h), b)
        out_file.write(str(cls_id) + " " + " ".join([str(a) for a in bb]) + '\n')
wd = getcwd()
print(wd)
for image_set in sets:
    if not os.path.exists('data/labels/'):
        os.makedirs('data/labels/')
    image_ids = open('data/ImageSets/%s.txt' % (image_set)).read().strip().split()
    list_file = open('data/%s.txt' % (image_set), 'w')
    for image_id in image_ids:
        list_file.write('data/images/%s.jpg\n' % (image_id))
        convert_annotation(image_id)
    list_file.close()

数据格式如下图：

其中images里面存放的是JPEGImages的全部图片。 

2.环境

（1）git clone https://github.com/ultralytics/yolov3.git

（2）pip install -U -r requirements.txt

（5）在项目根目录下新建weights文件夹，下载权重文件，将其放入weights文件夹中。

（6）测试：./darknet detect cfg/yolov3.cfg weights/yolov3.pt  data/samples/bus.jpg    或     ./darknet detector test cfg/coco.data cfg/yolov3.cfg yolov3.pt  data/samples/

3.训练模型

（1）下载。

（2）在data目录下新建**.name文件，存放你的数据集类别名称。本文用coco.names:

aeroplane
bicycle
bird
boat
bottle
bus
car
cat
chair
cow
diningtable
dog
horse
motorbike
person
pottedplant
sheep
sofa
train
tvmonitor

（3）在data目录下新建**.data文件，本文用coco.data：

classes = 20#类别数
train = data\2007_train.txt#voc_labels.py生成的训练集的位置
valid = data\2007_test.txt
names = data\coco.names
backup = backup\
（4） 更新cfg文件的classes，本文使用的classes=20。yolo上一卷积层的filters=3*（classes+5），其中5代表的是4个坐标+1个置信度。

（5）开始训练：python train.py --data data/coco.data --cfg cfg/yolov3.cfg  --weights weights/yolov3.pt

中断后，恢复训练：python train.py --data data/coco.data --cfg cfg/yolov3.cfg  --weights weights/yolov3.pt  --resume

注意：max_batches = 50200 ### 迭代次数

（6）测试：

python detect.py --cfg cfg/yolov3.cfg --weights/******.pt --source data/samples/file.jpg

• Image: --source file.jpg
• Video: --source file.mp4

（7）评估模型：

python test.py --data  data/coco.data  --cfg cfg/yolov3.cfg --weights weights/******.pt

（8）可视化：

Python -c from utils import utils; utils.plot_results()