您好,登录后才能下订单哦!
密码登录
登录注册
点击 登录注册 即表示同意《亿速云用户服务条款》
接触pytorch一天,发现pytorch上手的确比TensorFlow更快。可以更方便地实现用预训练的网络提特征。
以下是提取一张jpg图像的特征的程序:
# -*- coding: utf-8 -*- import os.path import torch import torch.nn as nn from torchvision import models, transforms from torch.autograd import Variable import numpy as np from PIL import Image features_dir = './features' img_path = "hymenoptera_data/train/ants/0013035.jpg" file_name = img_path.split('/')[-1] feature_path = os.path.join(features_dir, file_name + '.txt') transform1 = transforms.Compose([ transforms.Scale(256), transforms.CenterCrop(224), transforms.ToTensor() ] ) img = Image.open(img_path) img1 = transform1(img) #resnet18 = models.resnet18(pretrained = True) resnet50_feature_extractor = models.resnet50(pretrained = True) resnet50_feature_extractor.fc = nn.Linear(2048, 2048) torch.nn.init.eye(resnet50_feature_extractor.fc.weight) for param in resnet50_feature_extractor.parameters(): param.requires_grad = False #resnet152 = models.resnet152(pretrained = True) #densenet201 = models.densenet201(pretrained = True) x = Variable(torch.unsqueeze(img1, dim=0).float(), requires_grad=False) #y1 = resnet18(x) y = resnet50_feature_extractor(x) y = y.data.numpy() np.savetxt(feature_path, y, delimiter=',') #y3 = resnet152(x) #y4 = densenet201(x) y_ = np.loadtxt(feature_path, delimiter=',').reshape(1, 2048)
以下是提取一个文件夹下所有jpg、jpeg图像的程序:
# -*- coding: utf-8 -*- import os, torch, glob import numpy as np from torch.autograd import Variable from PIL import Image from torchvision import models, transforms import torch.nn as nn import shutil data_dir = './hymenoptera_data' features_dir = './features' shutil.copytree(data_dir, os.path.join(features_dir, data_dir[2:])) def extractor(img_path, saved_path, net, use_gpu): transform = transforms.Compose([ transforms.Scale(256), transforms.CenterCrop(224), transforms.ToTensor() ] ) img = Image.open(img_path) img = transform(img) x = Variable(torch.unsqueeze(img, dim=0).float(), requires_grad=False) if use_gpu: x = x.cuda() net = net.cuda() y = net(x).cpu() y = y.data.numpy() np.savetxt(saved_path, y, delimiter=',') if __name__ == '__main__': extensions = ['jpg', 'jpeg', 'JPG', 'JPEG'] files_list = [] sub_dirs = [x[0] for x in os.walk(data_dir) ] sub_dirs = sub_dirs[1:] for sub_dir in sub_dirs: for extention in extensions: file_glob = os.path.join(sub_dir, '*.' + extention) files_list.extend(glob.glob(file_glob)) resnet50_feature_extractor = models.resnet50(pretrained = True) resnet50_feature_extractor.fc = nn.Linear(2048, 2048) torch.nn.init.eye(resnet50_feature_extractor.fc.weight) for param in resnet50_feature_extractor.parameters(): param.requires_grad = False use_gpu = torch.cuda.is_available() for x_path in files_list: print(x_path) fx_path = os.path.join(features_dir, x_path[2:] + '.txt') extractor(x_path, fx_path, resnet50_feature_extractor, use_gpu)
另外最近发现一个很简单的提取不含FC层的网络的方法:
resnet = models.resnet152(pretrained=True) modules = list(resnet.children())[:-1] # delete the last fc layer. convnet = nn.Sequential(*modules)
另一种更简单的方法:
resnet = models.resnet152(pretrained=True) del resnet.fc
以上这篇pytorch实现用Resnet提取特征并保存为txt文件的方法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亿速云。
免责声明:本站发布的内容(图片、视频和文字)以原创、转载和分享为主,文章观点不代表本网站立场,如果涉及侵权请联系站长邮箱:is@yisu.com进行举报,并提供相关证据,一经查实,将立刻删除涉嫌侵权内容。