扫二维码与项目经理沟通
我们在微信上24小时期待你的声音
解答本文疑问/技术咨询/运营咨询/技术建议/互联网交流
这篇文章主要介绍了Pytorch中mean和std调查的示例分析,具有一定借鉴价值,感兴趣的朋友可以参考下,希望大家阅读完这篇文章之后大有收获,下面让小编带着大家一起了解一下。
创新互联公司主营罗山网站建设的网络公司,主营网站建设方案,重庆APP开发公司,罗山h5成都小程序开发搭建,罗山网站营销推广欢迎罗山等地区企业咨询如下所示:
# coding: utf-8 from __future__ import print_function import copy import click import cv2 import numpy as np import torch from torch.autograd import Variable from torchvision import models, transforms import matplotlib.pyplot as plt import load_caffemodel import scipy.io as sio # if model has LSTM # torch.backends.cudnn.enabled = False imgpath = 'D:/ck/files_detected_face224/' imgname = 'S055_002_00000025.png' # anger image_path = imgpath + imgname mean_file = [0.485, 0.456, 0.406] std_file = [0.229, 0.224, 0.225] raw_image = cv2.imread(image_path)[..., ::-1] print(raw_image.shape) raw_image = cv2.resize(raw_image, (224, ) * 2) image = transforms.Compose([ transforms.ToTensor(), transforms.Normalize( mean=mean_file, std =std_file, #mean = mean_file, #std = std_file, ) ])(raw_image).unsqueeze(0) print(image.shape) convert_image1 = image.numpy() convert_image1 = np.squeeze(convert_image1) # 3* 224 *224, C * H * W convert_image1 = convert_image1 * np.reshape(std_file,(3,1,1)) + np.reshape(mean_file,(3,1,1)) convert_image1 = np.transpose(convert_image1, (1,2,0)) # H * W * C print(convert_image1.shape) convert_image1 = convert_image1 * 255 diff = raw_image - convert_image1 err = np.max(diff) print(err) plt.imshow(np.uint8(convert_image1)) plt.show()
结论:
input_image = (raw_image / 255 - mean) ./ std
下面调查均值文件和方差文件是如何生成的:
mean_file = [0.485, 0.456, 0.406] std_file = [0.229, 0.224, 0.225]
# coding: utf-8 import matplotlib.pyplot as plt import argparse import os import numpy as np import torchvision import torchvision.transforms as transforms dataset_names = ('cifar10','cifar100','mnist') parser = argparse.ArgumentParser(description='PyTorchLab') parser.add_argument('-d', '--dataset', metavar='DATA', default='cifar10', choices=dataset_names, help='dataset to be used: ' + ' | '.join(dataset_names) + ' (default: cifar10)') args = parser.parse_args() data_dir = os.path.join('.', args.dataset) print(args.dataset) args.dataset = 'cifar10' if args.dataset == "cifar10": train_transform = transforms.Compose([transforms.ToTensor()]) train_set = torchvision.datasets.CIFAR10(root=data_dir, train=True, download=True, transform=train_transform) #print(vars(train_set)) print(train_set.train_data.shape) print(train_set.train_data.mean(axis=(0,1,2))/255) print(train_set.train_data.std(axis=(0,1,2))/255) # imshow image train_data = train_set.train_data ind = 100 img0 = train_data[ind,...] ## test channel number, in total , the correct channel is : RGB,not like BGR in caffe # error produce #b,g,r=cv2.split(img0) #img0=cv2.merge([r,g,b]) print(img0.shape) print(type(img0)) plt.imshow(img0) plt.show() # in ship in sea #img0 = cv2.resize(img0,(224,224)) #cv2.imshow('img0',img0) #cv2.waitKey() elif args.dataset == "cifar100": train_transform = transforms.Compose([transforms.ToTensor()]) train_set = torchvision.datasets.CIFAR100(root=data_dir, train=True, download=True, transform=train_transform) #print(vars(train_set)) print(train_set.train_data.shape) print(np.mean(train_set.train_data, axis=(0,1,2))/255) print(np.std(train_set.train_data, axis=(0,1,2))/255) elif args.dataset == "mnist": train_transform = transforms.Compose([transforms.ToTensor()]) train_set = torchvision.datasets.MNIST(root=data_dir, train=True, download=True, transform=train_transform) #print(vars(train_set)) print(list(train_set.train_data.size())) print(train_set.train_data.float().mean()/255) print(train_set.train_data.float().std()/255)
结果:
cifar10 Files already downloaded and verified (50000, 32, 32, 3) [ 0.49139968 0.48215841 0.44653091] [ 0.24703223 0.24348513 0.26158784] (32, 32, 3)
使用matlab检测是如何计算mean_file和std_file的:
% load cifar10 dataset data = load('cifar10_train_data.mat'); train_data = data.train_data; disp(size(train_data)); temp = mean(train_data,1); disp(size(temp)); train_data = double(train_data); % compute mean_file mean_val = mean(mean(mean(train_data,1),2),3)/255; % compute std_file temp1 = train_data(:,:,:,1); std_val1 = std(temp1(:))/255; temp2 = train_data(:,:,:,2); std_val2 = std(temp2(:))/255; temp3 = train_data(:,:,:,3); std_val3 = std(temp3(:))/255; mean_val = squeeze(mean_val); std_val = [std_val1, std_val2, std_val3]; disp(mean_val); disp(std_val); % result: mean_val: [0.4914, 0.4822, 0.4465] % std_val: [0.2470, 0.2435, 0.2616]
均值计算的过程也可以遵循标准差的计算过程。为 了简单,例如对于一个矩阵,所有元素的均值,等于两个方向上先后均值。所以会直接采用如下的形式:
mean_val = mean(mean(mean(train_data,1),2),3)/255;
标准差的计算是每一个通道的对所有样本的求标准差。然后再除以255。
感谢你能够认真阅读完这篇文章,希望小编分享的“Pytorch中mean和std调查的示例分析”这篇文章对大家有帮助,同时也希望大家多多支持创新互联成都网站设计公司,关注创新互联成都网站设计公司行业资讯频道,更多相关知识等着你来学习!
另外有需要云服务器可以了解下创新互联scvps.cn,海内外云服务器15元起步,三天无理由+7*72小时售后在线,公司持有idc许可证,提供“云服务器、裸金属服务器、网站设计器、香港服务器、美国服务器、虚拟主机、免备案服务器”等云主机租用服务以及企业上云的综合解决方案,具有“安全稳定、简单易用、服务可用性高、性价比高”等特点与优势,专为企业上云打造定制,能够满足用户丰富、多元化的应用场景需求。
我们在微信上24小时期待你的声音
解答本文疑问/技术咨询/运营咨询/技术建议/互联网交流