常见数据集格式转换

发表于 更新于 分类于 Waline： 本文字数： 131
coco（json）、yolo、mask深度学习数据集格式互相转换

经常遇到一些需要数据转换的场景，如coco格式、yolo格式（归一化记录在txt中）、mask图片，这里记录一下常用的格式转换

本篇内容：

mask转换为json（coco数据集的话自己再改一下）

python json_to_mask.py 文件夹或json文件 输出文件夹

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
import cv2
import os
import json
import sys


def func(file:str) -> dict:
    png = cv2.imread(file)
    gray = cv2.cvtColor(png, cv2.COLOR_BGR2GRAY)
    _, binary = cv2.threshold(gray,10,255,cv2.THRESH_BINARY)
    contours, _ = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

    dic = {"version": "5.0.1", "flags": {},"shapes":list(), "imagePath":os.path.basename(file),
            "imageHeight":png.shape[0], "imageWidth":png.shape[1]}
    for contour in contours:
        temp = list()
        for point in contour[2:]:
            if len(temp) > 1 and temp[-2][0] * temp[-2][1] * int(point[0][0]) * int(point[0][1]) != 0 and (int(point[0][0]) - temp[-2][0]) * (
                            temp[-1][1] - temp[-2][1]) == (int(point[0][1]) - temp[-2][1]) * (temp[-1][0] - temp[-1][0]):
                temp[-1][0] = int(point[0][0])
                temp[-1][1] = int(point[0][1])
            else:
                temp.append([int(point[0][0]), int(point[0][1])])
        dic["shapes"].append({"label": "result", "points":temp, "group_id": None,
                                "shape_type": "polygon", "flags": {}})

    return dic


if  __name__ == "__main__":

    if len(sys.argv) != 3:
        raise ValueError("mask文件或目录 输出路径")

    if os.path.isdir(sys.argv[1]):
        for file in os.listdir(sys.argv[1]):
            with open(os.path.join(sys.argv[2], os.path.splitext(file)[0]+".json"), mode='w', encoding="utf-8") as f:
                json.dump(func(os.path.join(sys.argv[1], file)), f)
    else:
        with open(os.path.join(sys.argv[2], os.path.splitext(os.path.basename(sys.argv[1]))[0]+".json"), mode='w', encoding="utf-8") as f:
            json.dump(func(sys.argv[1]), f)

可视化json

检查json格式是否正确，更改为json文件和对应的原图

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
import json
from PIL import Image, ImageDraw

# 读取原始图像
image = Image.open('images.png')

# 读取JSON实例坐标文件
with open('m04.json') as f:
    instances = json.load(f)
# 在原始图像上绘制实例
draw = ImageDraw.Draw(image)
for i,instance in enumerate(instances['shapes']):
    coords = instance['points']
    # 为每个实例指定不同颜色
    color = (255 * (i%3), 255 * ((i+1)%3), 255 * ((i+2)%3))
    for x, y in coords:
        draw.ellipse((x-2, y-2, x+2, y+2), fill=color)

image.show()
image.save('result.jpg')

json转化为YOLO格式

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
import os
import json
from PIL import Image, ImageDraw

json_folder = 'json/'
labels_folder = 'labels/'

# 遍历json文件夹
for json_file in os.listdir(json_folder):

    # 读取每个json文件
    with open(os.path.join(json_folder, json_file)) as f:
        instances = json.load(f)

    # 处理坐标
    all_coords = []
    for instance in instances['shapes']:
        coords = instance['points']
        instance_points = []
        for coord in coords:
            x, y = coord
            x /= instances['imageWidth']
            y /= instances['imageHeight']
            instance_points.append([x, y])
        all_coords.append(instance_points)

    # 写入labels文件夹
    label_file = os.path.join(labels_folder, json_file[1:-4] + 'txt')
    with open(label_file, 'w') as f:
        for coords in all_coords:
            if len(coords) == 0:
                continue
            f.write('0 ')
            for x, y in coords:
                f.write(f'{x} {y} ')
            f.write('\n')

参考：Mask图像与json标记文件的相互转换