即可啟動(dòng)Labelme。

三、Labelme標(biāo)注基本流程

四、Labelme標(biāo)注高級(jí)用法

五、樣例代碼

下面是一個(gè)簡(jiǎn)單的代碼實(shí)現(xiàn)，可以對(duì)圖片進(jìn)行標(biāo)注，并將結(jié)果保存成json文件。

import json
import os
import numpy as np
import base64
import cv2

from labelme import utils
from labelme._version import __version__ as labelme_version

def main():
    data = {}
    data['version'] = labelme_version
    data['flags'] = {}
    data['imagePath'] = 'test_image3.jpg'

    # 讀入圖片
    img_name = 'test_image3.jpg'
    img = cv2.imread(img_name)
    data['imageHeight'] = img.shape[0]
    data['imageWidth'] = img.shape[1]

    # 讀入標(biāo)注
    shapes = []
    with open('test_image3.json') as f:
        json_data = json.load(f)
    for i in json_data['shapes']:
        label = i['label']
        points = i['points']
        group_id = i['group_id']
        shape_type = i['shape_type']
        points_arr = np.array(points)
        shapes.append({'label': label, 'points': points_arr.tolist(), 
            'group_id': group_id, 'shape_type': shape_type, 
            'flags': {}})

    # 構(gòu)建結(jié)果
    data['shapes'] = shapes
    img_str = cv2.imencode('.jpg', img)[1].tostring()
    data['imageData'] = base64.b64encode(img_str).decode('utf-8')

    # 保存結(jié)果
    with open('result.json', 'w') as f:
        json.dump(data, f)