【目标检测】使用keras-yolov3训练自己的数据集详细流程

末蓝、 2023-06-23 08:59 6阅读 0赞

YOLOv3的具体的概念与流程这里不再介绍，下面直接讲解实现的步骤。

**计算机环境**：Win10 + Python3.6 + cuda9.0  
**IDE**：PyCharm + Anaconda  
**依赖**：

tensorflow-gpu 1.12.0
    keras-gpu 2.2.4
    opencv
    pillow
    numpy
    matplotlib

### 目录 ###

*  一 制作数据集
 *   *  1.1 LabelImg 标记图片
     *  1.2 VOC 数据集格式介绍
     *  1.3 转化成VOC数据集
     *  1.4 生成ImageSets
 *  二 代码修改
 *   *  2.1 修改\`voc\_annocation.py\`
     *  2.2 修改\`yolo3.cfg\`
     *  2.3 修改voc\_classes.txt
     *  2.4 修改train.py
     *  2.5 修改yolo.py

# 一 制作数据集 #

制作数据集的工程量较为巨大，如果你有几千张图片，就要一一为这些图片做标记。如果不是做实际的工程项目或是比赛，此步骤可以跳过，在互联网上直接下载已经标记好的数据集。

著名的数据集：  
1 [CIFAR-10 and CIFAR-100 datasets][] 已经处理好的数据，无法看到原图片。  
2 [PASCAL VOC][] 可直接查看原图片，数据集采用VOC格式

本文的数据集：[安全帽—百度网盘][Link 1] 采用VOC格式

## 1.1 LabelImg 标记图片 ##

标记图片就是对图片中的待识别目标进行标记，如果识别的目标时猫、狗，那就用方块标记出猫或狗，它们的标签分别为cat、dog。标记完成后，生成与文件名相同的`.xml`文件。  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1NlZV9TdGFy_size_16_color_FFFFFF_t_70]  
**LabelImg 安装及使用**  
标记图片所使用的软件为LabelImg，[图像标注工具labelImg安装教程及使用方法][labelImg]。

## 1.2 VOC 数据集格式介绍 ##

深度学习所使用的数据集格式大部分采用VOC格式，。所以我们也要将自己所标记的图片转变到VOC格式。

可在[PASCAL VOC][] 下载数据集查看VOC数据集。也可查看此文章：[VOC数据集格式介绍][VOC]，此文章以VOC2007为例，介绍了VOC的组成。

## 1.3 转化成VOC数据集 ##

标记好所有图片后，所有的`jpg`和`xml`均在一个目录下。我们新建一个文件夹`dataset`，在`dataset`目录下新建`JPEGImages`、`Annotations`和`ImageSets`三个文件夹。

将所有`jpg`文件放入`JPEGImages`中；  
将所有`xml`文件放入`Annotations`中；  
在`ImageSets`文件夹内新建`Main`备用。

## 1.4 生成ImageSets ##

使用代码`make_main_txt.py`生成`ImgeSets/Main`目录下的`train.txt`, `trainval.txt`, `val.txt`, `test.txt`。该代码存放的目录如下图，`make_main_txt.py`与`ImageSets`同级  
![在这里插入图片描述][20191227095758874.png]  
**代码**：`make_main_txt.py`

import os
    import random
    
    trainval_percent = 0.2 # test和val所占的比例，（1-trainval_percent）是训练集的比例
    train_percent = 0.8  # 其中在train时使用的是test所占trainval的比例
    
    xmlfilepath = 'Annotations'
    txtsavepath = 'ImageSets\Main'
    total_xml = os.listdir(xmlfilepath)
    
    num = len(total_xml)     # 图片数量
    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('ImageSets/Main/trainval.txt', 'w')
    ftest = open('ImageSets/Main/test.txt', 'w')
    ftrain = open('ImageSets/Main/train.txt', 'w')
    fval = open('ImageSets/Main/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()

运行之后，将产生下面这些文件。  
![在这里插入图片描述][2020020516421434.png]

# 二 代码修改 #

代码下载：[keras-yolo3][]  
权重下载：[yolov3.weights][] 并将权重放在keras-yolo3-master的文件夹下。

## 2.1 修改`voc_annocation.py` ##

该文件用于生成下图所示的文本文件，包括`train.txt`，`test.txt`，`val.txt`。  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1NlZV9TdGFy_size_16_color_FFFFFF_t_70 1]  
其中包含图片的目录，以及图片中标签位置数据。![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1NlZV9TdGFy_size_16_color_FFFFFF_t_70 2]  
对文件进行修改，修改结果如下，修改了类别（classes）、图片的打开目录和保存目录。

# -*- coding:utf-8 -*-
    import xml.etree.ElementTree as ET
    from os import getcwd
    
    sets = ['train', 'val', 'test']
    classes = ["person", "hat"]
    
    
    def convert_annotation(image_id, list_file):
        print(image_id)
        in_file = open('A05_helmet/Annotations/%s.xml' % image_id)
        tree = ET.parse(in_file)
        root = tree.getroot()
    
        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 = (int(xmlbox.find('xmin').text), int(xmlbox.find('ymin').text), int(xmlbox.find('xmax').text), int(xmlbox.find('ymax').text))
            list_file.write(" " + ",".join([str(a) for a in b]) + ',' + str(cls_id))
    
    wd = getcwd()
    
    for image_set in sets:
        image_ids = open('A05_helmet/ImageSets/Main/%s.txt' % image_set).read().strip().split()
        print(image_ids)
        list_file = open('%s.txt' % image_set, 'w')
    
        for image_id in image_ids:
            list_file.write('A05_helmet/JPEGImages/%s.jpg' % image_id)
            convert_annotation(image_id, list_file)
            list_file.write('\n')
        list_file.close()

## 2.2 修改`yolo3.cfg` ##

打开yolo3.cfg，搜索yolo(共三处)，每次均按下图修改。  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1NlZV9TdGFy_size_16_color_FFFFFF_t_70 3]  
filters：3\*(5+len(classes))

classes：训练的类别数

random：原来是1，显存小改为0

## 2.3 修改voc\_classes.txt ##

目录如下图所示，将`voc_classes.txt`的内容修改为为自己训练的类别。  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1NlZV9TdGFy_size_16_color_FFFFFF_t_70 4]

## 2.4 修改train.py ##

用下面代码直接替换原来的train.py代码。

""" Retrain the YOLO model for your own dataset. """
    import numpy as np
    import keras.backend as K
    from keras.layers import Input, Lambda
    from keras.models import Model
    from keras.callbacks import TensorBoard, ModelCheckpoint, EarlyStopping
     
    from yolo3.model import preprocess_true_boxes, yolo_body, tiny_yolo_body, yolo_loss
    from yolo3.utils import get_random_data
     
     
    def _main():
        annotation_path = 'train.txt'
        log_dir = 'logs/000/'
        classes_path = 'model_data/voc_classes.txt'
        anchors_path = 'model_data/yolo_anchors.txt'
        class_names = get_classes(classes_path)
        anchors = get_anchors(anchors_path)
        input_shape = (416,416) # multiple of 32, hw
        model = create_model(input_shape, anchors, len(class_names) )
        train(model, annotation_path, input_shape, anchors, len(class_names), log_dir=log_dir)
     
    def train(model, annotation_path, input_shape, anchors, num_classes, log_dir='logs/'):
        model.compile(optimizer='adam', loss={ 
            'yolo_loss': lambda y_true, y_pred: y_pred})
        logging = TensorBoard(log_dir=log_dir)
        checkpoint = ModelCheckpoint(log_dir + "ep{epoch:03d}-loss{loss:.3f}-val_loss{val_loss:.3f}.h5",
            monitor='val_loss', save_weights_only=True, save_best_only=True, period=1)
        batch_size = 10
        val_split = 0.1
        with open(annotation_path) as f:
            lines = f.readlines()
        np.random.shuffle(lines)
        num_val = int(len(lines)*val_split)
        num_train = len(lines) - num_val
        print('Train on {} samples, val on {} samples, with batch size {}.'.format(num_train, num_val, batch_size))
     
        model.fit_generator(data_generator_wrap(lines[:num_train], batch_size, input_shape, anchors, num_classes),
                steps_per_epoch=max(1, num_train//batch_size),
                validation_data=data_generator_wrap(lines[num_train:], batch_size, input_shape, anchors, num_classes),
                validation_steps=max(1, num_val//batch_size),
                epochs=500,
                initial_epoch=0)
        model.save_weights(log_dir + 'trained_weights.h5')
     
    def get_classes(classes_path):
        with open(classes_path) as f:
            class_names = f.readlines()
        class_names = [c.strip() for c in class_names]
        return class_names
     
    def get_anchors(anchors_path):
        with open(anchors_path) as f:
            anchors = f.readline()
        anchors = [float(x) for x in anchors.split(',')]
        return np.array(anchors).reshape(-1, 2)
     
    def create_model(input_shape, anchors, num_classes, load_pretrained=False, freeze_body=False,
                weights_path='model_data/yolo_weights.h5'):
        K.clear_session() # get a new session
        image_input = Input(shape=(None, None, 3))
        h, w = input_shape
        num_anchors = len(anchors)
        y_true = [Input(shape=(h//{ 0:32, 1:16, 2:8}[l], w//{ 0:32, 1:16, 2:8}[l], \
            num_anchors//3, num_classes+5)) for l in range(3)]
     
        model_body = yolo_body(image_input, num_anchors//3, num_classes)
        print('Create YOLOv3 model with {} anchors and {} classes.'.format(num_anchors, num_classes))
     
        if load_pretrained:
            model_body.load_weights(weights_path, by_name=True, skip_mismatch=True)
            print('Load weights {}.'.format(weights_path))
            if freeze_body:
                # Do not freeze 3 output layers.
                num = len(model_body.layers)-7
                for i in range(num): model_body.layers[i].trainable = False
                print('Freeze the first {} layers of total {} layers.'.format(num, len(model_body.layers)))
     
        model_loss = Lambda(yolo_loss, output_shape=(1,), name='yolo_loss',
            arguments={ 'anchors': anchors, 'num_classes': num_classes, 'ignore_thresh': 0.5})(
            [*model_body.output, *y_true])
        model = Model([model_body.input, *y_true], model_loss)
        return model
    def data_generator(annotation_lines, batch_size, input_shape, anchors, num_classes):
        n = len(annotation_lines)
        np.random.shuffle(annotation_lines)
        i = 0
        while True:
            image_data = []
            box_data = []
            for b in range(batch_size):
                i %= n
                image, box = get_random_data(annotation_lines[i], input_shape, random=True)
                image_data.append(image)
                box_data.append(box)
                i += 1
            image_data = np.array(image_data)
            box_data = np.array(box_data)
            y_true = preprocess_true_boxes(box_data, input_shape, anchors, num_classes)
            yield [image_data, *y_true], np.zeros(batch_size)
     
    def data_generator_wrap(annotation_lines, batch_size, input_shape, anchors, num_classes):
        n = len(annotation_lines)
        if n==0 or batch_size<=0: return None
        return data_generator(annotation_lines, batch_size, input_shape, anchors, num_classes)
     
    if __name__ == '__main__':
        _main()

替换完成后，需要**创建logs/000目录**。  
![在这里插入图片描述][20200205191321230.jpg]  
这个目录的作用就是存放自己的数据集训练得到的模型。不然程序运行到最后会因为找不到该路径而发生错误。生成的模型为`trained_weights.h5`。

## 2.5 修改yolo.py ##

修改yolo.py文件，修改如下两行目录。  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1NlZV9TdGFy_size_16_color_FFFFFF_t_70 5]

参考博客：[windows10+keras下的yolov3的快速使用及自己数据集的训练][windows10_keras_yolov3]

[CIFAR-10 and CIFAR-100 datasets]: http://www.cs.toronto.edu/~kriz/cifar.html
[PASCAL VOC]: http://host.robots.ox.ac.uk/pascal/VOC/
[Link 1]: https://pan.baidu.com/s/1UbFkGm4EppdAU660Vu7SdQ
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1NlZV9TdGFy_size_16_color_FFFFFF_t_70]: https://img-blog.csdnimg.cn/2019121812214721.png?x-ossprocess=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1NlZV9TdGFy,size_16,color_FFFFFF,t_70
[labelImg]: https://blog.csdn.net/See_Star/article/details/103324456
[VOC]: https://blog.csdn.net/See_Star/article/details/103546650
[20191227095758874.png]: https://img-blog.csdnimg.cn/20191227095758874.png
[2020020516421434.png]: https://img-blog.csdnimg.cn/2020020516421434.png
[keras-yolo3]: https://github.com/qqwweee/keras-yolo3
[yolov3.weights]: https://pjreddie.com/media/files/yolov3.weights
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1NlZV9TdGFy_size_16_color_FFFFFF_t_70 1]: https://img-blog.csdnimg.cn/20200205185539981.jpg?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1NlZV9TdGFy,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1NlZV9TdGFy_size_16_color_FFFFFF_t_70 2]: https://img-blog.csdnimg.cn/20200205164801780.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1NlZV9TdGFy,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1NlZV9TdGFy_size_16_color_FFFFFF_t_70 3]: https://img-blog.csdnimg.cn/20200205190357987.jpg?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1NlZV9TdGFy,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1NlZV9TdGFy_size_16_color_FFFFFF_t_70 4]: https://img-blog.csdnimg.cn/2020020519085949.jpg?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1NlZV9TdGFy,size_16,color_FFFFFF,t_70
[20200205191321230.jpg]: https://img-blog.csdnimg.cn/20200205191321230.jpg
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1NlZV9TdGFy_size_16_color_FFFFFF_t_70 5]: https://img-blog.csdnimg.cn/20200205191908637.jpg?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L1NlZV9TdGFy,size_16,color_FFFFFF,t_70
[windows10_keras_yolov3]: https://blog.csdn.net/u012746060/article/details/81183006