【数据增强可用】随机抠图，贴，想怎么旋转就怎么旋转

前提准备：

掩码制作见最后

原图一张
掩码图一张
例如：

代码

import numpy as np
import cv2
from PIL import Image

mask = Image.open(r'D:\Water_level_gauge\data\video_img\crop_img\stream0\mask\crop_5_18_stream0_day_0.jpg')  # 掩码图
img = Image.open(r'D:\Water_level_gauge\data\video_img\crop_img\stream0\img\crop_5_18_stream0_day_0.jpg')  # 原图

##########################  原图旋转  ###########################
resize_img = img.resize((int(img.size[0] * 0.3), int(img.size[1] * 0.3)))
resize_img = resize_img.rotate(45)

##########################  掩码旋转  ###########################
resize_mask = mask.resize((int(mask.size[0] * 0.3), int(mask.size[1] * 0.3)))
resize_mask = resize_mask.rotate(45)

##########################  从mask中随便找一个通道，cat到RGB后面做为掩码，最后转成RGBA  ###########################
res = np.concatenate((np.array(resize_img), np.array(resize_mask)[:, :, 0:1]), -1)
resize_img = Image.fromarray(res.astype('uint8'), mode='RGBA')

img.paste(resize_img)

img.show()

掩码制作

先用 精灵标注助手 将想要抠图的区域进行抠图，输出格式选择XML

from xml.dom.minidom import parse
import os
from PIL import Image, ImageDraw
import cv2
import numpy as np


# 将标注好的图片生成掩码图片
from xml.dom.minidom import parse
import os
from PIL import Image, ImageDraw
import cv2
import numpy as np


# 将标注好的图片生成掩码图片
def generate_mask(size, position):
    '''
    :param size: [w, h]
    :param position: list数据类型，点按顺序[x0,y0,x1,y1,x2,y2,x3,y3]
    :return: 返回PIL数据类型
    '''
    # 创建而背景图
    background = np.zeros([size[1], size[0], 3], dtype=np.uint8)

    # 创建mask掩码
    pts = np.array([[[position[0], position[1]], [position[2], position[3]],
                     [position[4], position[5]], [position[6], position[7]]]])

    cv2.fillPoly(background, [pts], (255, 255, 255))

    # 通道转换
    img_cv2 = cv2.cvtColor(background, cv2.COLOR_BGR2RGB)

    # 转成PIL数据格式
    img = Image.fromarray(img_cv2)
    # img.show()

    return img


def save_mask_png(xml_path, save_mask, is_save=False):
    for xml in os.listdir(xml_path):
        print(xml)

        xml_file = os.path.join(xml_path, xml)

        dom = parse(xml_file)  # 通过parse函数创建一个文件操作对象
        root = dom.documentElement  # 通过文本解析documentElement创建一个文本解析对象

        img_name = root.getElementsByTagName("path")[0].childNodes[0].data.split('\\')[-1]

        # 获取图片尺寸
        try:
            size = root.getElementsByTagName("size")[0]
            img_w = int(size.getElementsByTagName("width")[0].childNodes[0].data)
            img_h = int(size.getElementsByTagName("height")[0].childNodes[0].data)
            # print(img_w,img_h)

            # 创建背景图
            background = Image.new("RGB", (img_w, img_h), (0, 0, 0))
            draw = ImageDraw.Draw(background)

            object = root.getElementsByTagName("item")

            for box in object:
                # 如果超过4个点，就在后面继续添加即可
                x1 = int(box.getElementsByTagName("x1")[0].childNodes[0].data.split('.')[0])
                y1 = int(box.getElementsByTagName("y1")[0].childNodes[0].data.split('.')[0])
                x2 = int(box.getElementsByTagName("x2")[0].childNodes[0].data.split('.')[0])
                y2 = int(box.getElementsByTagName("y2")[0].childNodes[0].data.split('.')[0])
                x3 = int(box.getElementsByTagName("x3")[0].childNodes[0].data.split('.')[0])
                y3 = int(box.getElementsByTagName("y3")[0].childNodes[0].data.split('.')[0])
                x4 = int(box.getElementsByTagName("x4")[0].childNodes[0].data.split('.')[0])
                y4 = int(box.getElementsByTagName("y4")[0].childNodes[0].data.split('.')[0])
                mask_img = generate_mask((img_w, img_h), [x1, y1, x2, y2, x3, y3, x4, y4])
                if is_save:
                    mask_img.save(os.path.join(save_mask, img_name))
                else:
                    mask_img.show()
        except:
            x = 0
            # print('此图不生成样本')


if __name__ == '__main__':
    xml_path = r'D:\coding\Water_level_gauge\data\output\outputs'
    save_mask = r'D:\coding\Water_level_gauge\data\mask_png'
    save_mask_png(xml_path, save_mask,is_save=False)