labelme数据增强，原数据集图片使用方框标注（非圆点），纯python代码实现。

先看结果吧

然后进入正题，展示一下数据集格式

再展示一下修改后的数据集格式（每一张图片扩充了五张）

这是数据增强的代码，记得按自己文件的路径和命名进行少量修改

import cv2
import json
import numpy as np
import base64
import io
from PIL import Image
import random

# 读取JSON标注文件
def data_build():
    #共有45张图片
    for i in range(1, 46):
        #每张图片随机裁剪五下
        for j in range(1, 6):
            json_file_path = f'dataset/{str(i).zfill(2)}.json'
            # 读取JSON文件
            with open(json_file_path, 'r') as f:
                data = json.load(f)

            image_data = data['imageData']
            image_data = base64.b64decode(image_data)
            # 创建新的JSON数据对象
            new_data = {
                "version": data["version"],
                "flags": data["flags"],
                "shapes": [],
                "imagePath": data["imagePath"],
                "imageData": [],
                "imageHeight": 416,
                "imageWidth": 416
            }


            # 将图像数据转换为NumPy数组
            image_array = np.frombuffer(image_data, dtype=np.uint8)

            # 解码图像数组为OpenCV图像
            image = cv2.imdecode(image_array, cv2.IMREAD_COLOR)

            # 获取原始图像尺寸
            height, width, _ = image.shape

            # 设置目标裁剪尺寸
            target_size = 416

            # 随机选择裁剪位置
            x = random.randint(0, width - target_size)
            y = random.randint(0, height - target_size)
            print('x的范围为',x,'--',x+416)
            print('y的范围为',y,'--',y+416)

            # 裁剪图像
            cropped_image = image[y:y+target_size, x:x+target_size]

            # 调整裁剪后的图像大小为416x416
            resized_image = cv2.resize(cropped_image, (target_size, target_size))

            # 将图像转换为Base64编码
            _, buffer = cv2.imencode('.jpg', resized_image)
            base64_cropped_image = base64.b64encode(buffer).decode()

            new_data['imageData'] = base64_cropped_image


            # 遍历每个形状
            for shape in data['shapes']:
                points = shape['points']
                print("points:",points)
                # 检查每个点的坐标

                x1, y1 = points[0]
                x2, y2 = points[1]
                x_ = (x1+x2)/2
                y_ = (y1+y2)/2

                if x_>x and x_<x+416 and y_>y and y_<y+416:
                    print("找到了一个点,x为：",x1,',y为',y1)
                    x1 = x1-x
                    x2 = x2-x
                    y1 = y1-y
                    y2 = y2-y
                    points = [[x1,y1],[x2,y2]]
                    shape['points'] = points
                    new_data['shapes'].append(shape)



            # 将新的JSON数据保存到新的文件中
            with open(f'{i}_{j}.json', 'w') as f:
                json.dump(new_data, f, indent=4)

data_build()

 裁剪完毕，读取看看

import json
import cv2
import base64
import numpy as np

# 读取JSON文件
with open('dataset_build/26_3.json', 'r') as f:
    data = json.load(f)

# 读取图像路径和图像数据
image_path = data["imagePath"]
image_data = data["imageData"]

# 解码图像数据
image_data = base64.b64decode(image_data)

# 将图像数据转换为NumPy数组
image_array = np.frombuffer(image_data, dtype=np.uint8)

# 解码图像数组为OpenCV图像
image = cv2.imdecode(image_array, cv2.IMREAD_COLOR)

# 遍历每个形状
for shape in data['shapes']:
    label = shape['label']
    points = shape['points']

    # 绘制方框
    x1, y1 = points[0]
    x2, y2 = points[1]
    cv2.rectangle(image, (int(x1), int(y1)), (int(x2), int(y2)), (0, 0, 255), 2)

# 显示图像
cv2.imshow('Image with Annotations', image)
cv2.waitKey(0)
cv2.destroyAllWindows()

 结果：