模拟数据

最新推荐文章于 2024-09-09 09:30:03 发布

被自己蠢哭了

最新推荐文章于 2024-09-09 09:30:03 发布

阅读量646

点赞数 1

分类专栏：深度学习

本文链接：https://blog.csdn.net/jicong44/article/details/84958457

版权

深度学习专栏收录该内容

34 篇文章 0 订阅

订阅专栏

针对警告标志进行模拟数据的生成

从标准库中截取标准图案，去掉白色背景

# 将空白处设置为零
img_dir = './all_moni_img'
save_dir = './all_moni_img_bgblack'
imgs_name = os.listdir(img_dir)
for img_name in imgs_name:
    img = cv2.imread(os.path.join(img_dir,img_name))
    img = np.array(img)
    
    bool_ar = img<250
    img =img*bool_ar
    
    img = cv2.resize(img,(48,48))
    cv2.imwrite(os.path.join(save_dir,img_name),img)

对上面得到的结果进行数据增广

import cv2,os
import imgaug as ia
from imgaug import augmenters as iaa
import numpy as np

ia.seed(1)
seq = iaa.Sequential([
#    iaa.Fliplr(1.0), # 按概率水平翻转
    iaa.Crop(percent=(0, 0.1)), # 随机裁剪
    iaa.GaussianBlur(sigma=(1., 1.0)), # 模糊处理
#    iaa.ContrastNormalization((1., 1.5)), # 进行对比度调整
#    iaa.AdditiveGaussianNoise(loc=0, scale=(0.0, 0.005*255)), # 高斯噪声,对每个图片都在N(0,s)（均值为零，方差在0
                                                               # 到0.005*255之间随机取）的正太分布上进行一次采样。
    iaa.Multiply((0.1,1.5)), # 亮度调节
    iaa.Affine(
        scale={"x": (0.8, 1.2), "y": (0.8, 1.2)},
        translate_percent={"x": (-0.1, 0.1), "y": (-0.1, 0.1)},
        rotate=(-5, 5),
#        shear=(-8, 8)
    ) # 仿射变换 scale表示尺度，translate_percent: 平移比例,0表示不平移,0.5表示平移50%。
      # rotate: 平移角度,0~360度之间,正负也可以表示方向，shear: 错切的程度,0~360度之间,正负表示方向。
      # 详细参考：https://blog.csdn.net/a8039974/article/details/82791732
], random_order=False) # apply augmenters in random order


def mk_dir(pa):
    if not os.path.exists(pa):
        os.mkdir(pa)

if __name__ == '__main__':
    imgs_name = os.listdir('/home/tf/China_TSR_test_new_net/test_teacher_idea/all_moni_img_bgblack')
    for img_name in imgs_name:
        mk_dir('/home/tf/China_TSR_test_new_net/test_teacher_idea/aug_sou_img/'+img_name[:-4])
        img = cv2.imread('/home/tf/China_TSR_test_new_net/test_teacher_idea/all_moni_img_bgblack/'+img_name)
        for i in range(50):    
            img_aug = seq.augment_image(img)
            cv2.imwrite('/home/tf/China_TSR_test_new_net/test_teacher_idea/aug_sou_img/%s/au_%d'%(img_name[:-4],i)+img_name,img_aug)

从三张图片中随机截取背景，将交通标志贴上去

# 从实际场景中裁背景，将图片贴上去。
imgs_name = ['000028.jpg','000029.jpg','000030.jpg']

for img_name in imgs_name:
    img = cv2.imread(img_name)
    fils_name = os.listdir('./aug_sou_img')
    for fil_name in fils_name:
        print('img_name:',img_name,'fil_name:',fil_name)
        mk_dir('./res/'+fil_name)
        ts_imgs_name = os.listdir('./aug_sou_img/'+fil_name)
        for ts_img_name in ts_imgs_name:
            ts_img = cv2.imread('./aug_sou_img/'+fil_name+'/'+ts_img_name)
            for gh in range(5):
                ran = int(random.random()*120)
                while ran<60:
                    ran = int(random.random()*120)
            
                x1 = int(np.random.random()*1000)
                y1 = int(np.random.random()*700)
                bg = img[y1:y1+ran,x1:x1+ran,:]
                bg = cv2.resize(bg,(52,52))
                sou_bg = np.copy(bg)
                bg[int(bg.shape[0]/2-24):int(bg.shape[0]/2+24),int(bg.shape[0]/2-24):int(bg.shape[0]/2+24),:] = ts_img
                h,w,c = bg.shape
                for i in range(h):
                    for j in range(w):
                        for k in range(c):
                            if(bg[i][j][k]==0):
                                bg[i][j][k] = sou_bg[i][j][k]
                cv2.imwrite('./res/'+fil_name+'/%s_%s_%d.png'%(img_name[:-4],ts_img_name[:-4],gh),bg)