python多张不同大小的图片按照间距合并/填充、新掩膜计算

先是题目，应该是要再def merge(images, annotations_file):填的

参考代码（第一问）

balloon.py

# -*- coding: utf-8 -*- 
import cv2
import numpy as np
from opencv.coding_pentest.opencvjson import allll
"""
需求：
    1、在尺寸为 2048 * 2048 的黑色背景图片中，依次填充 images 中的气球图片，气球图
       片的间隔至少大于 20 个像素，填充完成并保存，如果填充溢出，可保存多张图片。。
    2、annotations.json 中对应每一个气球的标注的掩膜坐标信息，请根据 annotations 
       中的格式生成新的 json 标注文件，并确保新生成的 json 标注文件里的掩膜坐标信息对
       应需求 1 中保存的图片。
    3、请验证需求 2 中新生成 json 文件掩膜标注信息的正确性。
"""
def merge(images, annotations_file):
    pass
def merge():
    allll()
    pass
merge()

opencvjson.py第二第三问就是一个线性变换，第116行去掉注释可验证第三问

# -*- coding: utf-8 -*-
import cv2
import os
import numpy as np
import json


def cv_show(name, img, k=0, width=512, height=512):
    cv2.namedWindow(name, cv2.WINDOW_NORMAL)
    cv2.resizeWindow(name, width, height)
    cv2.imshow(name, img)
    cv2.waitKey(k)


def letter_box(image, dim, interval=0):
    h, w = dim
    h -= interval * 2
    w -= interval * 2
    height, width = image.shape[0], image.shape[1]
    ratio = min(w/width, h/height)
    nw = int(width * ratio)
    nh = int(height * ratio)
    dh = (h-nh)//2
    dw = (w-nw)//2
    resized_image = cv2.resize(image, (nw, nh))
    output = np.zeros((dim[0], dim[1], 3), dtype=np.uint8)
    output[dh+interval:dh+interval+nh, dw+interval:dw+interval+nw, :] = resized_image
    return output, ratio, dh, dw

def merge1(a,num):
# 输入图片
    imgName = 'merge_0.jpg'
    if(num>15):
        imgName = 'merge_1.jpg'
    img = cv2.imread(imgName)

    # 展示原图
    #cv2.imshow("img", img)
    # 创建掩膜
    mask = np.zeros(img.shape[:2], dtype=np.uint8)
    #a=[[27, 175], [510, 152], [519, 471], [24, 474]]
    #a=[[156, 160, 164, 169, 172, 180, 188, 196, 204, 209, 212, 223, 236, 262, 284, 304, 324, 351, 373, 391, 413, 435, 454, 461, 472, 479, 486, 493, 499, 499, 498, 491, 484, 476, 467, 456, 437, 407, 386, 362, 341, 316, 288, 264, 236, 216, 196, 180, 166, 153, 141, 133, 122, 113, 109, 109, 114, 118, 125, 136, 147],[405, 411, 417, 422, 426, 433, 441, 447, 452, 455, 457, 460, 463, 464, 462, 456, 449, 436, 422, 409, 389, 366, 341, 330, 311, 295, 275, 246, 211, 180, 157, 129, 108, 92, 77, 63, 45, 24, 14, 6, 1, 0, 2, 7, 19, 30, 46, 60, 79, 98, 119, 132, 162, 198, 228, 257, 292, 318, 343, 369, 392]]
    #a=np.transpose(a)

    polygon = np.array(a, np.int32) # 坐标为顺时针方向
    cv2.fillConvexPoly(mask, polygon, (255, 255, 255))
    image = cv2.add(img, np.zeros(np.shape(img), dtype=np.uint8), mask=mask)
    # 展示掩膜图片
    #cv2.imshow("mask", mask)
    # 展示添加掩膜效果图片
    cv2.imshow("image", image)

    cv2.waitKey()
    cv2.destroyAllWindows()
    #cv2.imwrite('01.png',mask)



def allll():
    root_path = './images/'
    names = os.listdir(root_path)
    imgs = [cv2.imread(os.path.join(root_path, name)) for name in names]

    merge_size = (2048, 2048)   # 背景图尺寸
    rows = 4                    # 行数设定
    columns = 4                 # 列数设定
    interval = 20               # 图片间的最小间隔

    # 计算子图需要resize的尺寸
    assert merge_size[0] % columns == 0 and merge_size[1] % rows == 0, '%s or %s is not multiples of %s' % (rows, columns, merge_size)
    h = merge_size[0] // columns
    w = merge_size[1] // rows

    # 计算需要几张背景图
    if len(names) <= rows*columns:
        num = 1
    else:
        if len(names) % (rows*columns) == 0:
            num = len(names)//(rows*columns)
        else:
            num = len(names)//(rows*columns)+1
    canvas = np.zeros((num, merge_size[0], merge_size[1], 3), dtype=np.uint8)

    f = open('annotations.json', 'r')
    content = f.read()
    json_in = json.loads(content)
    print(type(json_in))

    with open('1.txt', 'a') as f:  # 'a'表示append,即在原来文件内容后继续写数据（不清楚原有数据）
        f.write("{")

    #print(json_in['%s' % (names[0])]['all_points_x'])
    content_s="{"
    # 合并图片
    cnt = 0
    for n in range(len(names)):
        print(names[n])
        for k in range(num):
            for j in range(columns):
                for i in range(rows):
                    if cnt == len(names):
                        break
                    canvas[k][j*h:(j+1)*h, i*w:(i+1)*w, :], ratio, dh, dw = letter_box(imgs[cnt+n//(rows*columns)], (h, w), interval=interval)

                    # TODO: 读入json数据，通过ratio, dw, dh, interval计算合并后的新位置
                    # ratio是压缩比例a, b是每张图的左上角xy坐标, y = ax + b
                    # dw=90,dh=0
                    # b0x=20i+90,b0y=20i+0

                    a=[(json_in['%s'%(names[cnt])]['all_points_x']),(json_in['%s'%(names[cnt])]['all_points_y'])]
                    a=np.transpose(a)
                    for lo in range(a.shape[0]):
                        a[lo][0]=a[lo][0]*ratio+20+512*(cnt%16%4)+dw
                        a[lo][1]=a[lo][1]*ratio+20+512*(cnt%16//4)+dh
#将下一行取消注释可验证第三问
                    #merge1(a,cnt)
                    a = np.transpose(a)
                    print(a[0])
                    content_s +=('"'+names[cnt]+'":')

                    dict_out1 = dict(name="polygon",all_points_x=a[0].tolist(),all_points_y=a[1].tolist())
                    print(dict_out1)
                    content_s += (str(dict_out1))+","
                    # cv_show('black', canvas[k])
                    cnt += 1

            cv2.imwrite('merge_%s.jpg' % k, canvas[k])
    print('cnt: ', cnt)
    content_s =content_s[:-1]
    content_s += "}"
    content_s=content_s.replace("\'",'\"')
    with open('1.txt', 'r+') as f:
        f.writelines(content_s)
        f.close()
    os.rename("1.txt", "1.json")