【矩池云】YOLOv3~YOLOv5训练红外小目标数据集

一、数据集准备

数据集下载地址：https://github.com/YimianDai/sirst

1. 需要将数据集转换为YOLO所需要的txt格式

参考链接：https://github.com/pprp/voc2007_for_yolo_torch

1.1 检测图片及其xml文件

import os, shutil

def checkPngXml(dir1, dir2, dir3, is_move=True):
    """
    dir1 是图片所在文件夹
    dir2 是标注文件所在文件夹
    dir3 是创建的，如果图片没有对应的xml文件，那就将图片放入dir3
    is_move 是确认是否进行移动，否则只进行打印
    """
    if not os.path.exists(dir3):
        os.mkdir(dir3)
    cnt = 0
    for file in os.listdir(dir1):
        f_name,f_ext = file.split(".")
        if not os.path.exists(os.path.join(dir2, f_name+".xml")):
            print(f_name)
            if is_move:
                cnt += 1
                shutil.move(os.path.join(dir1,file), os.path.join(dir3, file))
    if cnt > 0:
        print("有%d个文件不符合要求，已打印。"%(cnt))
    else:
        print("所有图片和对应的xml文件都是一一对应的。")

if __name__ == "__main__":
    dir1 = r"dataset/images/images"   # 修改为自己的图片路径
    dir2 = r"dataset/masks/masks"     # 修改为自己的图片路径
    dir3 = r"dataset/Allempty"        # 修改为自己的图片路径
    checkPngXml(dir1, dir2, dir3, False)

1.2 划分训练集

import os
import random
import os, fnmatch
 

trainval_percent = 0.8
train_percent = 0.8

xmlfilepath = r"dataset/masks/masks"
txtsavepath = r"dataset"
total_xml = fnmatch.filter(os.listdir(xmlfilepath), '*.xml')
print(total_xml)

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

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

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

1.3 转为txt标签

# -*- coding: utf-8 -*-
"""
需要修改的地方：
1. sets中替换为自己的数据集
2. classes中替换为自己的类别
3. 将本文件放到VOC2007目录下
4. 直接开始运行
"""

import xml.etree.ElementTree as ET
import pickle
import os
from os import listdir, getcwd
from os.path import join
sets=[('2007', 'train'), ('2007', 'val'), ('2007', 'test')]  #替换为自己的数据集
classes = ["Target"]     #修改为自己的类别

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(year, image_id):
    in_file = open('dataset/masks/masks/%s.xml'%(image_id))  #将数据集放于当前目录下
    out_file = open('dataset/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)
    print(w,h)
    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)
        print(cls_id)
        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)
        print(bb)
        out_file.write(str(cls_id) + " " + " ".join([str(a) for a in bb]) + '\n')
wd = getcwd()
for year, image_set in sets:
    if not os.path.exists('dataset/labels/'):
        os.makedirs('dataset/labels/')
    image_ids = open('dataset/%s.txt'%(image_set)).read().strip().split()
    list_file = open('%s_%s.txt'%(year, image_set), 'w')
    for image_id in image_ids:
        list_file.write('dataset/images/images/%s.png\n'%(image_id))
        convert_annotation(year, image_id)
    list_file.close()
# os.system("cat 2007_train.txt 2007_val.txt > train.txt")     #修改为自己的数据集用作训练

1.4 构造数据集

import os, shutil

"""
需要满足以下条件：
1. 在JPEGImages中准备好图片
2. 在labels中准备好labels
3. 创建好如下所示的文件目录：
    - images
        - train2014
        - val2014
    - labels(由于voc格式中有labels文件夹，所以重命名为label)
        - train2014
        - val2014
"""


def make_for_torch_yolov3(dir_image,
                                 dir_label,
                                 dir1_train,
                                 dir1_val,
                                 dir2_train,
                                 dir2_val,
                                 main_trainval,
                                 main_test):
    if not os.path.exists(dir1_train):
        os.mkdir(dir1_train)
    if not os.path.exists(dir1_val):
        os.mkdir(dir1_val)
    if not os.path.exists(dir2_train):
        os.mkdir(dir2_train)
    if not os.path.exists(dir2_val):
        os.mkdir(dir2_val)

    with open(main_trainval, "r") as f1:
        for line in f1:
            print(line[:-1])
            # print(os.path.join(dir_image, line[:-1]+".png"), os.path.join(dir1_train, line[:-1]+".png"))
            shutil.copy(os.path.join(dir_image, line[:-1]+".png"),
                        os.path.join(dir1_train, line[:-1]+".png"))
            shutil.copy(os.path.join(dir_label, line[:-1]+".txt"),
                        os.path.join(dir2_train, line[:-1]+".txt"))


    with open(main_test, "r") as f2:
        for line in f2:
            print(line[:-1])
            shutil.copy(os.path.join(dir_image, line[:-1]+".png"),
                        os.path.join(dir1_val, line[:-1]+".png"))
            shutil.copy(os.path.join(dir_label, line[:-1]+".txt"),
                        os.path.join(dir2_val, line[:-1]+".txt"))

if __name__ == "__main__":
    '''
    https://github.com/ultralytics/yolov3
    这个pytorch版本的数据集组织
    - images
        - train2014 # dir1_train
        - val2014 # dir1_val
    - labels
        - train2014 # dir2_train
        - val2014 # dir2_val
    trainval.txt, test.txt 是由create_main.py构建的
    '''

    dir_image = r"dataset/images/images"
    dir_label = r"dataset/labels"

    dir1_train = r"dataset/image/train2014"
    dir1_val = r"dataset/image/val2014"

    dir2_train = r"dataset/label/train2014"
    dir2_val = r"dataset/label/val2014"

    main_trainval = r"dataset/trainval.txt"
    main_test = r"dataset/test.txt"

    make_for_torch_yolov3(dir_image,
                            dir_label,
                            dir1_train,
                            dir1_val,
                            dir2_train,
                            dir2_val,
                            main_trainval,
                            main_test)

最终数据集格式如下：

 2. 构造训练所需要的数据集

根据以上数据集 需要单独构建一个datasets文件夹，存放标签和图像，具体格式如下：

可以参考该链接：https://github.com/ultralytics/yolov5/issues/7389

3. 构建数据配置文档，需要注意 YOLOv5目录需要和datasets目录同级。

命名为hongwai.yaml

# YOLOv3 🚀 by Ultralytics, AGPL-3.0 license
# COCO 2017 dataset http://cocodataset.org by Microsoft
# Example usage: python train.py --data coco.yaml
# parent
# ├── yolov5
# └── datasets
#     └── coco  ← downloads here (20.1 GB)


# Train/val/test sets as 1) dir: path/to/imgs, 2) file: path/to/imgs.txt, or 3) list: [path/to/imgs1, path/to/imgs2, ..]
path: ../datasets/coco # dataset root dir
train: images/train2014  # train images (relative to 'path') 118287 images
val: images/val2014  # val images (relative to 'path') 5000 images
test: images/val2014  # 20288 of 40670 images, submit to https://competitions.codalab.org/competitions/20794

# Classes
nc: 1
names:
  0: Target

 二、矩池云配置环境

1. 租用环境

 2、 配置环境，缺啥配啥，耐心解决问题

参考命令：

git clone https://github.com/ultralytics/yolov3.git

cd yolov3
pip install -r requirements.txt

也许训练过程中还会报错找不到module，根据module名字，使用pip安装即可

三、训练

YOLOv5训练命令：

python train.py --data data/hongwai.yaml --weights '' --cfg yolov5s.yaml --img 640 --device 0

YOLOv3训练命令：

python train.py --data data/hongwai.yaml --weights '' --cfg yolov3.yaml --img 640 --device 0

训练结果部分展示：

 

 

 四、文件夹检测

执行命令：

python detect.py --weights runs/train/exp10/weights/best.pt --source dataset/image/val2014

结果保存位置： 

 

 

【创造不易，需要指导做该项目的可以联系】