通过labelme的json文件实现对图片的批量裁剪
通过labelme的json文件实现对图片的批量裁剪
工具安装
首先我们需要在conda的终端下载labelme，然后打开labelme对图片进行抠图，将一张图需要抠的指定元素打好标签。具体操作很简单需要进一步了解的可以点击这里
抠图程序编写
首先我们将每张图对应的json文件命名与原图一样，以便于后续操作，并将全部json文件统一放到统一的文件夹下面，我这里的文件夹是picturejson
json文件的具体样式为：
{
"version": "5.0.1",
"flags": {},
"shapes": [
{
"label": "yang",
"points": [
[
34.09803921568627,
200.0
],
[
79.19607843137254,
162.7450980392157
],
[
200.76470588235293,
166.66666666666666
],
[
418.4117647058823,
162.7450980392157
],
[
447.82352941176475,
245.09803921568627
],
[
426.2549019607843,
327.45098039215685
],
[
439.98039215686276,
411.7647058823529
],
[
706.6470588235294,
325.4901960784314
],
[
718.4117647058823,
1788.235294117647
],
[
32.13725490196077,
1796.078431372549
]
],
"group_id": null,
"shape_type": "polygon",
"flags": {}
},
{
"label": "yang",
"points": [
[
1043.9019607843138,
168.62745098039215
],
[
1392.921568627451,
164.70588235294116
],
[
1432.1372549019607,
254.90196078431373
],
[
1406.6470588235293,
350.98039215686276
],
[
1438.0196078431372,
1780.392156862745
],
[
841.9411764705882,
1813.7254901960785
],
[
743.9019607843137,
1156.862745098039
],
[
745.8627450980392,
890.1960784313725
],
[
857.6274509803922,
774.5098039215686
],
[
994.8823529411764,
760.7843137254902
],
[
983.1176470588234,
200.0
]
],
"group_id": null,
"shape_type": "polygon",
"flags": {}
},
{
"label": "yang",
"points": [
[
2416.450980392157,
164.70588235294116
],
[
2851.7450980392155,
168.62745098039215
],
[
2857.627450980392,
1782.3529411764705
],
[
2189.0,
1792.156862745098
],
[
2153.705882352941,
905.8823529411765
],
[
2120.372549019608,
190.19607843137254
]
],
"group_id": null,
"shape_type": "polygon",
"flags": {}
}
],
"imagePath": "00001P.jpg",
"imageData": null,
"imageHeight": 1901,
"imageWidth": 2874
}
注意
我们在抠图的时候不要把图片的imagedata给保存，这个对勾给取消，不然数据乱码，json文件会识别不出来。
然后我们还需要对应的图片的数据集以便于我们可以通过json文件找到指定的图片并对指定的图片进行操作，这里要保证图片可以找到。
具体程序编码
import os
import cv2
import json
import numpy as np
jsonPath = "./picturejson/"
jpgDirPath = "./dataset/"
jsonName = os.listdir(jsonPath)
for json1 in jsonName:
img = cv2.imread(jpgDirPath + json1.split(".")[0] + ".jpg")
path = jsonPath +json1
print(path)
j = 0
with open(path, 'r', encoding='utf-8') as f:
load_dict = json.load(f)
dic_data = load_dict["shapes"]
for i in dic_data:
pts = np.array(i["points"])
pts = pts.astype(np.int64)
rect = cv2.boundingRect(pts)
x, y, w, h = rect
croped = img[y:y + h, x:x + w].copy()
pts = pts - pts.min(axis=0)
mask = np.zeros(croped.shape[:2], np.uint8)
cv2.drawContours(mask, [pts], -1, (255, 255, 255), -1, cv2.LINE_AA)
dst = cv2.bitwise_and(croped, croped, mask=mask)
bg = np.ones_like(croped, np.uint8) * 255
cv2.bitwise_not(bg, bg, mask=mask)
dst2 = bg + dst
Save_File = "./cutout/"
Name_File = i['label']
category = load_dict['imagePath']
if "P" in category.split('.')[0] or "p" in category.split('.')[0]:
save_path = os.path.join(Save_File,Name_File+'/psave', load_dict['imagePath'].split(".")[0]+"_"+str(j) + '.jpg')
else:
save_path = os.path.join(Save_File, Name_File + '/',load_dict['imagePath'].split(".")[0] + "_" + str(j) + '.jpg')
if os.path.exists(Save_File + Name_File):
cv2.imwrite(save_path, dst2)
else:
os.makedirs(Save_File + Name_File + "/psave")
j += 1
结果
程序会在当前文件夹下进行创建一个cutout文件夹并且同时会在cutout文件夹下对数据的类别进行分类存储。
