MMdetection训练voc数据集

1、安装MMdetection
查看官方的安装文档：https://mmdetection.readthedocs.io/zh_CN/latest/get_started.html#id2



从git上面下载方法，需要注意最后这个命令。
2、准备voc格式数据集
MMdetection支持coco和voc数据集两种格式，也可以自定义数据集。最好使用的是coco格式，我先用的是voc格式，所以介绍一下voc格式。
首先需要了解一下voc数据集的格式：




自己数据集转换为voc数据集格式的代码：

import os
from glob import glob
import cv2
from lxml.etree import Element, SubElement, tostring
import numpy as np


# YOLO格式的txt转VOC格式的xml
def convert(img, box):
    
    name, x, y, w, h = box
    img_w = img.shape[0]
    img_h = img.shape[1]
    x = float(x) * img_h
    w = float(w) * img_h
    y = float(y) * img_w
    h = float(h) * img_w
    x = (x * 2 - w) / 2
    y = (y * 2 - h) / 2
    # print(name)
    # print((x,y,w,h))
    # exit()
    return name, x, y, w, h


# 单个文件转换
def txt_xml(img_name):
    clas = []
    img = cv2.imdecode(np.fromfile(os.path.join(img_path, img_name), dtype=np.uint8), cv2.IMREAD_UNCHANGED)
    # imh, imw = img.shape[0:2]  
    imh=800
    imw=800
    # print(img_name[:-4])
    # exit()
    txt_img = os.path.join(txt_path, img_name[:-4] + '.txt')  #txt_path TOD/labels/test    img_name[:-4] 014348
    with open(txt_img, "r") as f:
        for line in f.readlines():
            line = line.strip('\n')
            list = line.split(" ")
            while '' in list:
                list.remove('')
            # print(list)
            # exit()
            list = convert(img, list)
            clas.append(list)
        # print(clas)
        # exit()    
    node_root = Element('annotation')
    node_folder = SubElement(node_root, 'folder')
    node_folder.text = 'imge'
    node_filename = SubElement(node_root, 'filename')
    node_filename.text = img_name
    node_filepath = SubElement(node_root, 'path')
    node_filepath.text = img_path + "\\" + img_name
    node_size = SubElement(node_root, 'size')
    node_width = SubElement(node_size, 'width')
    node_width.text = str(imw)
    node_height = SubElement(node_size, 'height')
    node_height.text = str(imh)
    node_depth = SubElement(node_size, 'depth')
    node_depth.text = '3'
    for i in range(len(clas)):
        node_object = SubElement(node_root, 'object')
        node_name = SubElement(node_object, 'name')
        node_name.text = str(classes[int(clas[i][0])])
        node_pose = SubElement(node_object, 'pose')
        node_pose.text = "Unspecified"
        node_truncated = SubElement(node_object, 'truncated')
        node_truncated.text = "truncated"
        node_difficult = SubElement(node_object, 'difficult')
        node_difficult.text = '0'
        node_bndbox = SubElement(node_object, 'bndbox')
        node_xmin = SubElement(node_bndbox, 'xmin')
        node_xmin.text = str(clas[i][1])
        node_ymin = SubElement(node_bndbox, 'ymin')
        node_ymin.text = str(clas[i][2])
        node_xmax = SubElement(node_bndbox, 'xmax')
        node_xmax.text = str(clas[i][1] + clas[i][3])
        node_ymax = SubElement(node_bndbox, 'ymax')
        node_ymax.text = str(clas[i][2] + clas[i][4])
    xml = tostring(node_root, pretty_print=True)  # 格式化显示，该换行的换行
    img_newxml = os.path.join(xml_path, img_name[:-4] + '.xml')
    file_object = open(img_newxml, 'wb')
    file_object.write(xml)
    file_object.close()


# 批量转换
def generate_label_file():
    # 获取所有图片
    imgs = []
    jpg_imgs = glob("{}/*.jpg".format(img_path))
    # png_imgs = glob("{}/*.png".format(img_path))
    imgs.extend(jpg_imgs)
    # imgs.extend(png_imgs)
    for img in imgs:
        # print(img.split(os.sep)[-1])
        # exit()
        txt_xml(img.split(os.sep)[-1])  #img->TOD/images/test/014348.jpg  img.split(os.sep)[-1]->014348.jpg


if __name__ == "__main__":
    # 图片路径、txt文件路径、xml文件保存路径
    img_path = r"TOD/JPEGImages"
    txt_path = r"TOD/Annotations"
    xml_path = r"TOD/Annotations2"

    # 类别列表
    classes = ["a", "b"]
    # 执行转换
    generate_label_file()

3、修改配置文件
（1）修改数据集配置文件：

（2）修改模型：这里的模型是你需要跑的实验的模型

4、开始训练