Posted in Python onJuly 01, 2020
1. 用于分类模型:
import numpy as np import scipy.misc import cv2 import os # DF1 path = "/home/pi/工作/predict1/" npy_list = os.listdir(path) save_path = "/home/pi/predict1_img/" if not os.path.exists(save_path): os.mkdir(save_path) for i in range(0, len(npy_list)): print(i) print(npy_list[i]) npy_full_path = os.path.join(path, npy_list[i]) img = np.load(npy_full_path) # load进来 save_full_path = os.path.join(save_path, npy_list[i][:-4]) scipy.misc.imsave(save_full_path, img) # 保存
2. 用于分割模型
""" 将数据集随机分成训练集、测试集 传入参数: ratio = 0.7 # 训练样本比例 path = "/home/pi/20190701_0705" # 数据路径 new_path = "/home/pi/20190701_0705_new2" # 保存路径 使用方法: temp = Generate_Train_and_Test(path, new_path, ratio) temp.splict_data() """ import random import os import cv2 def makeDir(path): try: if not os.path.exists(path): if not os.path.isfile(path): # os.mkdir(path) os.makedirs(path) return 0 else: return 1 except Exception as e: print(str(e)) return -2 class Generate_Train_and_Test: def __init__(self, path, new_path, ratio): if not os.path.exists(new_path): makeDir(new_path) self.path = path self.new_path = new_path self.ratio = ratio self.train_sample_path = os.path.join(new_path, "train") self.test_sample_path = os.path.join(new_path, "test") makeDir(self.train_sample_path) makeDir(self.test_sample_path) def splict_data(self): class_names = os.listdir(self.path) # 类别:bg and ng10 for name in class_names: print("process class name=%s" % name) tmp_class_name = os.path.join(self.path, name) save_train_class_name = os.path.join(self.train_sample_path, name) save_test_class_name = os.path.join(self.test_sample_path, name) makeDir(save_train_class_name) makeDir(save_test_class_name) if os.path.isdir(tmp_class_name): image_names = os.listdir(tmp_class_name) # 其中一个类别的所有图像 image_names = [f for f in image_names if not f.endswith('_mask.png')] total = len(image_names) # 1, 打乱当前类中所有图像 random.shuffle(image_names) # 2, 从当前类(ng)中,取前面的图像作为train data train_temp = int(self.ratio * total) # 打乱后,取前面作为train_data for i in range(0, train_temp): print(i, image_names[i]) temp_img_name = os.path.join(tmp_class_name, image_names[i]) train_image = cv2.imread(temp_img_name) temp_label_name = os.path.join(tmp_class_name, image_names[i][:-4] + '_mask.png') train_label = cv2.imread(temp_label_name) save_train_img_name = os.path.join(save_train_class_name, image_names[i]) cv2.imwrite(save_train_img_name, train_image) save_train_label_name = os.path.join(save_train_class_name, image_names[i][:-4] + '_mask.png') cv2.imwrite(save_train_label_name, train_label) # 3, 从当前类(bg)中,取后面的图像作为test data for i in range(train_temp, total): print(i, image_names[i]) test_img_name = os.path.join(tmp_class_name, image_names[i]) test_image = cv2.imread(test_img_name) test_label_name = os.path.join(tmp_class_name, image_names[i][:-4] + '_mask.png') test_label = cv2.imread(test_label_name) save_test_img_name = os.path.join(save_test_class_name, image_names[i]) cv2.imwrite(save_test_img_name, test_image) save_test_label_name = os.path.join(save_test_class_name, image_names[i][:-4] + '_mask.png') cv2.imwrite(save_test_label_name, test_label) ratio = 0.7 # 训练样本比例 path = "/home/pi/工作/20190712_splict" # 数据路径 new_path = "/home/pi/工作/20190712_splict_new3" # 保存路径 temp = Generate_Train_and_Test(path, new_path, ratio) temp.splict_data()
补充知识:python把由图片组成的文件夹转换为.npy文件
由于深度神经网络的需要,我要将一个里面全是.png格式的图片的文件夹转换为一个.npy文件,即将一个图片文件夹转换成一个.npy文件。
具体思路为:
若已知文件夹中图片数量,可生成一个三维数组,第一维表示图片数量,后两维表示一张图片的尺寸;
利用np.save()函数将生成的三维数组保存成一个.npy文件
import numpy as np import imageio import os os.chdir('E:/RegistrationCode/papercode/datasets/mri_2d_test') #切换python工作路径到你要操作的图片文件夹,mri_2d_test为我的图片文件夹 a=np.ones((190,192,160)) #利用np.ones()函数生成一个三维数组,当然也可用np.zeros,此数组的每个元素a[i]保存一张图片 i=0 for filename in os.listdir(r"E:/RegistrationCode/papercode/datasets/mri_2d_test"): #使用os.listdir()获取该文件夹下每一张图片的名字 im=imageio.imread(filename) a[i]=im i=i+1 if(i==190): #190为文件夹中的图片数量 break np.save('你要保存的.npy文件所在路径及名字',a)
以上这篇使用npy转image图像并保存的实例就是小编分享给大家的全部内容了,希望能给大家一个参考,也希望大家多多支持三水点靠木。
使用npy转image图像并保存的实例
- Author -
Mr.Q声明:登载此文出于传递更多信息之目的,并不意味着赞同其观点或证实其描述。
Reply on: @reply_date@
@reply_contents@