PyTorch深度学习（19）网络结构VGGNet

 VGGNet

网络结构

分类数据&＃xff1a;将数据集中10%的图像为验证集Validation&＃xff0c;其他为训练集Train

import os
from shutil import copy, rmtree
import randomdef mk_file(file_path: str):if os.path.exists(file_path):# 如果文件夹存在&＃xff0c;则先删除原文件夹在重新创建rmtree(file_path)os.makedirs(file_path)def main():# 保证随机可复现random.seed(0)# 将数据集中10%的数据划分到验证集中split_rate &＃61; 0.1# 指向你解压后的flower_photos文件夹cwd &＃61; os.getcwd()data_root &＃61; os.path.join(cwd, "data_set")origin_flower_path &＃61; os.path.join(data_root, "flower_photos")assert os.path.exists(origin_flower_path), "path &＃39;{}&＃39; does not exist.".format(origin_flower_path)# 获取类型flower_class &＃61; [cla for cla in os.listdir(origin_flower_path)if os.path.isdir(os.path.join(origin_flower_path, cla))]# 建立保存训练集的文件夹train_root &＃61; os.path.join(data_root, "train")mk_file(train_root)for cla in flower_class:# 建立每个类别对应的文件夹mk_file(os.path.join(train_root, cla))# 建立保存验证集的文件夹val_root &＃61; os.path.join(data_root, "val")mk_file(val_root)for cla in flower_class:# 建立每个类别对应的文件夹mk_file(os.path.join(val_root, cla))for cla in flower_class:cla_path &＃61; os.path.join(origin_flower_path, cla)images &＃61; os.listdir(cla_path)num &＃61; len(images)# 随机采样验证集的索引 10%数量为验证集eval_index &＃61; random.sample(images, k&＃61;int(num*split_rate))for index, image in enumerate(images):if image in eval_index:# 将分配至验证集中的文件复制到相应目录image_path &＃61; os.path.join(cla_path, image)new_path &＃61; os.path.join(val_root, cla)copy(image_path, new_path)else:# 将分配至训练集中的文件复制到相应目录image_path &＃61; os.path.join(cla_path, image)new_path &＃61; os.path.join(train_root, cla)copy(image_path, new_path)print("\r[{}] processing [{}/{}]".format(cla, index&＃43;1, num), end&＃61;"") # processing barprint()print("processing done!")if __name__ &＃61;&＃61; &＃39;__main__&＃39;:main()

VGG Class

import torch.nn as nn
import torch# official pretrain weights
model_urls &＃61; {&＃39;vgg11&＃39;: &＃39;https://download.pytorch.org/models/vgg11-bbd30ac9.pth&＃39;,&＃39;vgg13&＃39;: &＃39;https://download.pytorch.org/models/vgg13-c768596a.pth&＃39;,&＃39;vgg16&＃39;: &＃39;https://download.pytorch.org/models/vgg16-397923af.pth&＃39;,&＃39;vgg19&＃39;: &＃39;https://download.pytorch.org/models/vgg19-dcbb9e9d.pth&＃39;
}class VGG(nn.Module):def __init__(self, feature, class_num&＃61;1000, init_weights&＃61;False):super(VGG, self).__init__()self.feature &＃61; featureself.classifier &＃61; nn.Sequential(nn.Linear(512*7*7, 4096),nn.ReLU(True),nn.Dropout(p&＃61;0.5),nn.Linear(4096, 4096),nn.ReLU(True),nn.Dropout(p&＃61;0.5),nn.Linear(4096, class_num))if init_weights:self._initialize_weights()def forward(self, x):# N × 3 × 224 × 224x &＃61; self.feature(x)# N × 512 × 7 × 7x &＃61; torch.flatten(x, start_dim&＃61;1)# N × 512 × 7 × 7x &＃61; self.classifier(x)return xdef _initialize_weights(self):for m in self.modules():if isinstance(m, nn.Conv2d):# nn.init.kaiming_normal_(m.weight, mode&＃61;&＃39;fan_out&＃39;, nonlinearity&＃61;&＃39;relu&＃39;)nn.init.xavier_uniform_(m.weight) # 初始化卷积核权重if m.bias is not None:nn.init.constant_(m.bias, 0) # 将偏置初始化为0elif isinstance(m, nn.Linear):nn.init.xavier_uniform_(m.weight)# nn.init.normal_(m.weight, 0, 0.01)nn.init.constant_(m.bias, 0)def make_feature(cfg: list):layers &＃61; []in_channels &＃61; 3for v in cfg:if v &＃61;&＃61; &＃39;M&＃39;:layers &＃43;&＃61; [nn.MaxPool2d(kernel_size&＃61;2, stride&＃61;2)]else:conv2d &＃61; nn.Conv2d(in_channels, v, kernel_size&＃61;3, padding&＃61;1)layers &＃43;&＃61; [conv2d, nn.ReLU(True)]in_channels &＃61; vreturn nn.Sequential(*layers)cfgs &＃61; {&＃39;vgg11&＃39;: [64, &＃39;M&＃39;, 128, &＃39;M&＃39;, 256, 256, &＃39;M&＃39;, 512, 512, &＃39;M&＃39;, 512, 512, &＃39;M&＃39;],&＃39;vgg13&＃39;: [64, 64, &＃39;M&＃39;, 128, 128, &＃39;M&＃39;, 256, 256, &＃39;M&＃39;, 512, 512, &＃39;M&＃39;, 512, 512, &＃39;M&＃39;],&＃39;vgg16&＃39;: [64, 64, &＃39;M&＃39;, 128, 128, &＃39;M&＃39;, 256, 256, 256, &＃39;M&＃39;, 512, 512, 512, &＃39;M&＃39;, 512, 512, 512, &＃39;M&＃39;],&＃39;vgg19&＃39;: [64, 64, &＃39;M&＃39;, 128, 128, &＃39;M&＃39;, 256, 256, 256, 256, &＃39;M&＃39;, 512, 512, 512, 512, &＃39;M&＃39;, 512, 512, 512, 512, &＃39;M&＃39;]
}# 实例化VGG模型
def vgg(model_name&＃61;"vgg16", **kwargs):try:cfg &＃61; cfgs[model_name]except:print("warning: model number {} not in cfgs dict!".format(model_name))exit(-1)model &＃61; VGG(make_feature(cfg), **kwargs)return model


Training Phase

import torch
import torchvision
import os
import json
import sys
import torch.nn as nn
import torch.optim as optim
import torchvision.transforms as transforms
import torch.utils.data.dataloader
from VGGNet import vgg
from tqdm import tqdm # python进度条def main():data_transform &＃61; {"train": transforms.Compose([transforms.RandomResizedCrop(224), # 随机裁剪transforms.RandomHorizontalFlip(), # 随机水平翻转transforms.ToTensor(),transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]),"val": transforms.Compose([transforms.Resize((224, 224)),transforms.ToTensor(),transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])}DEVICE &＃61; torch.device(&＃39;cuda:0&＃39; if torch.cuda.is_available() else &＃39;cpu&＃39;)print(DEVICE)data_root &＃61; os.path.abspath(os.path.join(os.getcwd(), "")) # get data root pathimage_path &＃61; data_root &＃43; "/data_set/" # flower data set pathassert os.path.exists(image_path), "{} path does not exist.".format(image_path)train_dataset &＃61; torchvision.datasets.ImageFolder(root&＃61;os.path.join(image_path, "train"),transform&＃61;data_transform["train"])train_num &＃61; len(train_dataset)# {&＃39;daisy&＃39;:0, &＃39;dandelion&＃39;:1, &＃39;roses&＃39;:2, &＃39;sunflower&＃39;:3, &＃39;tulips&＃39;:4}flower_list &＃61; train_dataset.class_to_idxcla_dict &＃61; dict((val, key) for key, val in flower_list.items())# write dict into json filejson_str &＃61; json.dumps(cla_dict, indent&＃61;4)with open(&＃39;class_indices.json&＃39;, &＃39;w&＃39;) as json_file:json_file.write(json_str)batch_size &＃61; 2nw &＃61; min([os.cpu_count(), batch_size if batch_size > 1 else 0, 8]) # number of workersprint(&＃39;Using {} dataloader workers every process&＃39;.format(nw))train_loader &＃61; torch.utils.data.DataLoader(train_dataset, batch_size&＃61;batch_size, shuffle&＃61;True, num_workers&＃61;0)validate_dataset &＃61; torchvision.datasets.ImageFolder(root&＃61;image_path &＃43; "val", transform&＃61;data_transform["val"])val_num &＃61; len(validate_dataset)validate_loader &＃61; torch.utils.data.DataLoader(validate_dataset, batch_size&＃61;batch_size, shuffle&＃61;False, num_workers&＃61;0)print("using {} images for training, {} images for validation.".format(train_num, val_num))# test_data_iter &＃61; iter(validate_loader)# test_image, test_label &＃61; test_data_iter.next()model_name &＃61; "vgg16"net &＃61; vgg(model_name&＃61;model_name, class_num&＃61;5, init_weights&＃61;True)net.to(DEVICE)loss_function &＃61; nn.CrossEntropyLoss()optimizer &＃61; optim.Adam(net.parameters(), lr&＃61;0.0001)epochs &＃61; 30best_acc &＃61; 0.0save_path &＃61; &＃39;./{}Net.pth&＃39;.format(model_name)train_steps &＃61; len(train_loader)for epoch in range(epochs):# trainnet.train()running_loss &＃61; 0.0train_bar &＃61; tqdm(train_loader, file&＃61;sys.stdout)for step, data in enumerate(train_bar):images, labels &＃61; dataoptimizer.zero_grad()outputs &＃61; net(images.to(DEVICE))loss &＃61; loss_function(outputs, labels.to(DEVICE))loss.backward()optimizer.step()# print statisticsrunning_loss &＃43;&＃61; loss.item()train_bar.desc &＃61; "train epoch[{}/{}] loss:{:.3f}".format(epoch &＃43; 1, epochs, loss)# validatenet.eval()acc &＃61; 0.0 # accumulate accurate number / epochwith torch.no_grad():val_bar &＃61; tqdm(validate_loader, file&＃61;sys.stdout)for val_data in val_bar:val_images, val_labels &＃61; val_dataoutputs &＃61; net(val_images.to(DEVICE))predict_y &＃61; torch.max(outputs, dim&＃61;1)[1]acc &＃43;&＃61; torch.eq(predict_y, val_labels.to(DEVICE)).sum().item()val_accurate &＃61; acc / val_numprint(&＃39;[epoch %d] train_loss: %.3f val_accuracy: %.3f&＃39; % (epoch &＃43; 1, running_loss / train_steps, val_accurate))if val_accurate > best_acc:best_acc &＃61; val_accuratetorch.save(net.state_dict(), save_path)print(&＃39;Finished Training&＃39;)if __name__ &＃61;&＃61; &＃39;__main__&＃39;:main()


Test Phase

使用一张图片来识别是那种类型

import os
import jsonimport torch
from PIL import Image
from torchvision import transforms
import matplotlib.pyplot as pltfrom model import vggdef main():device &＃61; torch.device("cuda:0" if torch.cuda.is_available() else "cpu")data_transform &＃61; transforms.Compose([transforms.Resize((224, 224)),transforms.ToTensor(),transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])# load imageimg_path &＃61; "../tulip.jpg"assert os.path.exists(img_path), "file: &＃39;{}&＃39; dose not exist.".format(img_path)img &＃61; Image.open(img_path)plt.imshow(img)# [N, C, H, W]img &＃61; data_transform(img)# expand batch dimensionimg &＃61; torch.unsqueeze(img, dim&＃61;0)# read class_indictjson_path &＃61; &＃39;./class_indices.json&＃39;assert os.path.exists(json_path), "file: &＃39;{}&＃39; dose not exist.".format(json_path)json_file &＃61; open(json_path, "r")class_indict &＃61; json.load(json_file)# create modelmodel &＃61; vgg(model_name&＃61;"vgg16", num_classes&＃61;5).to(device)# load model weightsweights_path &＃61; "./vgg16Net.pth"assert os.path.exists(weights_path), "file: &＃39;{}&＃39; dose not exist.".format(weights_path)model.load_state_dict(torch.load(weights_path, map_location&＃61;device))model.eval()with torch.no_grad():# predict classoutput &＃61; torch.squeeze(model(img.to(device))).cpu()predict &＃61; torch.softmax(output, dim&＃61;0)predict_cla &＃61; torch.argmax(predict).numpy()print_res &＃61; "class: {} prob: {:.3}".format(class_indict[str(predict_cla)],predict[predict_cla].numpy())plt.title(print_res)for i in range(len(predict)):print("class: {:10} prob: {:.3}".format(class_indict[str(i)],predict[i].numpy()))plt.show()if __name__ &＃61;&＃61; &＃39;__main__&＃39;:main()