【数据集制作】各种格式互转xml2json、image2xml、xml2txt

参考链接

• 相互转换的脚本：https://github.com/guchengxi1994/mask2json
• 批量分割mask转xml检测框

暂存自己的代码（拉跨，没有太大参考价值）

1.image2txtone

只转单个类别

import os
import cv2


# 传入图像地址，统计某张照片中的所有像素值
def countpixel(rootdir):
    assert os.path.exists(rootdir), f'{rootdir} not exist.'
    arr = []  # 查看了tunnelmask的像素值
    for file in os.listdir(rootdir):
        img = cv2.imread(os.path.join(rootdir, file))
        height, width = img.shape[:-1]
        for h in range(height):
            for w in range(width):
                pixel = img[h, w].tolist()
                if pixel not in arr:
                    arr.append(pixel)
    print(f'arr={arr}')
    return arr


# 从一张图中找到[1, 1, 1]出现的x_min x_max y_min y_max
def position_lwd(imagepath, cls_list):
    image = cv2.imread(imagepath)
    height, width = image.shape[:-1]

    x = [[]]
    y = [[]]
    cnt = 0  # 实例个数
    for h in range(height):
        for w in range(width):
            pixel = image[h, w].tolist()
            if pixel != [0, 0, 0]:  # 如果不为背景
                if w not in x:
                    x.append(w)
                if h not in y:
                    y.append(h)

    # 找出x_min x_max y_min y_max
    x_min = min(x)
    x_max = max(x)
    y_min = min(y)
    y_max = max(y)
    return (x_min, x_max, y_min, y_max)


# 归一化处理
def convert(size, box):
    # 该函数将xmin,ymin,xmax,ymax转为x,y,w,h中心点坐标和宽高
    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(rootdir, out_dir, cls_list):
    assert os.path.exists(rootdir), f'{rootdir} not exist.'
    if not os.path.exists(out_dir):
        os.makedirs(out_dir)

    filelist = os.listdir(rootdir)
    for file in filelist:
        filename = os.path.splitext(file)[0]
        file_path = os.path.join(rootdir, file)
        file_cv = cv2.imread(file_path)
        h, w = file_cv.shape[:-1]

        # 输出标签的地址
        out_path = os.path.join(out_dir, filename + '.txt')
        out_file = open(out_path, 'w')

        # 开始转换
        box = position_lwd(file_path, cls_list)  # 多类别转换
        bb = convert([w, h], box)
        out_file.write(str(0) + " " + " ".join([str(round(a, 3)) for a in bb]) + '\n')


if __name__ == '__main__':
    rootdir = r'D:\SoftWareInstallMenu\JetBrains\PycharmProjects\tunnel_datadeal\makesegdata\11'
    out_dir = r'D:\SoftWareInstallMenu\JetBrains\PycharmProjects\tunnel_datadeal\makesegdata\11_save'
    cls_list = countpixel(rootdir)  # 所有类别，第零位是[0,0,0]背景不算做类别，所以cls_id要减1
    convert_annotation(rootdir, out_dir, cls_list)

2.image2txt

转多类别

import os
import cv2


# 传入图像地址，统计某张照片中的所有像素值
def countpixel(rootdir):
    assert os.path.exists(rootdir), f'{rootdir} not exist.'
    arr = []  # 查看了tunnelmask的像素值
    for file in os.listdir(rootdir):
        img = cv2.imread(os.path.join(rootdir, file))
        height, width = img.shape[:-1]
        for h in range(height):
            for w in range(width):
                pixel = img[h, w].tolist()
                if pixel not in arr:
                    arr.append(pixel)
    print(f'arr={arr}')
    return arr


# 从一张图中找到[1, 1, 1]出现的x_min x_max y_min y_max
def position_lwd(imagepath, cls_list):
    image = cv2.imread(imagepath)
    height, width = image.shape[:-1]
    val = []
    x = []
    y = []
    for h in range(height):
        for w in range(width):
            pixel = image[h, w].tolist()
            if pixel != [0, 0, 0]:  # 如果不为背景
                if pixel not in val:
                    val.append(pixel)
                    x.append([])
                    y.append([])
                index = val.index(pixel)
                x[index].append(w)
                y[index].append(h)

    # 找出每一个类别的x_min x_max y_min y_max
    boxs = dict()
    for i in range(len(val)):
        cls_id = cls_list.index(val[i]) - 1  # 类别编号
        boxs[cls_id] = []
        boxs[cls_id].append(min(x[i]))
        boxs[cls_id].append(max(x[i]))
        boxs[cls_id].append(min(y[i]))
        boxs[cls_id].append(max(y[i]))

    return boxs


# 归一化处理
def convert(size, box):
    # 该函数将xmin,ymin,xmax,ymax转为x,y,w,h中心点坐标和宽高
    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(rootdir, out_dir, cls_list):
    assert os.path.exists(rootdir), f'{rootdir} not exist.'
    if not os.path.exists(out_dir):
        os.makedirs(out_dir)

    filelist = os.listdir(rootdir)
    for file in filelist:
        filename = os.path.splitext(file)[0]
        file_path = os.path.join(rootdir, file)
        file_cv = cv2.imread(file_path)
        h, w = file_cv.shape[:-1]

        # 输出标签的地址
        out_path = os.path.join(out_dir, filename + '.txt')
        out_file = open(out_path, 'w')

        # 开始转换
        boxs = position_lwd(file_path, cls_list)  # 多类别转换
        for cls_id, box in boxs.items():
            bb = convert([w, h], box)
            out_file.write(str(cls_id) + " " + " ".join([str(round(a, 3)) for a in bb]) + '\n')


if __name__ == '__main__':
    rootdir = r'D:\SoftWareInstallMenu\JetBrains\PycharmProjects\yolov5ds_lwd\datadeal\makesegdata\tunnel_mask'
    out_dir = r'D:\SoftWareInstallMenu\JetBrains\PycharmProjects\yolov5ds_lwd\datadeal\makesegdata\tunnel_mask_out'
    cls_list = countpixel(rootdir)  # 所有类别，第零位是[0,0,0]背景不算做类别，所以cls_id要减1
    convert_annotation(rootdir, out_dir, cls_list)