Yolov5语义分割使用

1.标注

2. 训练

3.预测

1. 标注

我这里用Labelme标注，标注完是Json格式，要把Json转Yolov5能识别的txt格式。在Python中转换。代码如下：

 

import os
import cv2
import json
import numpy as np


def txt_write(x, img_x, img_y, txt):
    data = x['points']
    n = 1
    for x in data:
        for i in x:
            if n % 2 == 0:
                txt.write(' ' + str(round(i / img_x, 6)))
                n += 1
            else:
                txt.write(' ' + str(round(i / img_y, 6)))
                n += 1
    txt.write('\n')


def json2txt(json_path, save_path):
    txt = open(save_path, 'w')
    with open(json_path, "r") as f:
        data = f.read()
    data = json.loads(data)
    img_x = data['imageHeight']
    img_y = data['imageWidth']
    shapes = data['shapes']

    for x in shapes:
        # print(x['label'])
        # 此处面向不同分类，需要改动下面的标签值，如果是多分类，那么需要增加新的if
        # 只是单分类的话，可以直接去掉if，把里面的模块拿出来用

        if x['label'] == 'bad':
            txt.write('0')
            txt_write(x, img_x, img_y, txt)
    txt.close()


# 单文件测试
# save_dir = "/workspace/" #文件路径
# name = 'test'
# save_path = save_dir + name + '.txt' # 也可以是.doc
# json_path = '/json/65161.json'
# json2txt(json_path,save_path)

# 文件夹
json_dir = 'D:\\yolov5-7.0\\datasets\\xt\\'
save_dir = 'D:\\yolov5-7.0\\datasets\\txt\\'
files = os.listdir(json_dir)
os.makedirs(save_dir, exist_ok=True)
num = 1
for file in files:
    name = file[0:-5]
    json_path = json_dir + '/' + name + '.json'
    save_path = save_dir + '/' + name + '.txt'
    json2txt(json_path, save_path)
    print(num, '/', len(files), ':', name)
    num += 1

2 . 训练参数设置

到segment这个文件夹下找这个train.py.    我这里命名是data/bvn-seg.yaml， 可以到Data目录下修改

data/bvn-seg.yaml

训练和评估图片路径

labels路径

3. 预测