|
| 1 | + |
| 2 | +import torch |
| 3 | +from torch import nn |
| 4 | +import torch.nn.functional as F |
| 5 | +import torchvision |
| 6 | +import torchvision.transforms as transforms |
| 7 | +from torchvision import models |
| 8 | +from torchvision.datasets import ImageFolder |
| 9 | +from datetime import datetime |
| 10 | + |
| 11 | + |
| 12 | +def get_acc(output, label): |
| 13 | + total = output.shape[0] |
| 14 | + _, pred_label = output.max(1) |
| 15 | + num_correct = (pred_label == label).sum().item() |
| 16 | + return num_correct / total |
| 17 | + |
| 18 | + |
| 19 | +def train(net, train_data, valid_data, num_epochs, optimizer, criterion): |
| 20 | + |
| 21 | + prev_time = datetime.now() |
| 22 | + for epoch in range(num_epochs): |
| 23 | + train_loss = 0 |
| 24 | + train_acc = 0 |
| 25 | + net = net.train() |
| 26 | + for im, label in train_data: |
| 27 | + im = im.to(device) # (bs, 3, h, w) |
| 28 | + label = label.to(device) # (bs, h, w) |
| 29 | + # forward |
| 30 | + output = net(im) |
| 31 | + loss = criterion(output, label) |
| 32 | + # backward |
| 33 | + optimizer.zero_grad() |
| 34 | + loss.backward() |
| 35 | + optimizer.step() |
| 36 | + |
| 37 | + train_loss += loss.item() |
| 38 | + train_acc += get_acc(output, label) |
| 39 | + |
| 40 | + cur_time = datetime.now() |
| 41 | + h, remainder = divmod((cur_time - prev_time).seconds, 3600) |
| 42 | + m, s = divmod(remainder, 60) |
| 43 | + time_str = "Time %02d:%02d:%02d" % (h, m, s) |
| 44 | + if valid_data is not None: |
| 45 | + valid_loss = 0 |
| 46 | + valid_acc = 0 |
| 47 | + net = net.eval() |
| 48 | + for im, label in valid_data: |
| 49 | + im = im.to(device) # (bs, 3, h, w) |
| 50 | + label = label.to(device) # (bs, h, w) |
| 51 | + output = net(im) |
| 52 | + loss = criterion(output, label) |
| 53 | + valid_loss += loss.item() |
| 54 | + valid_acc += get_acc(output, label) |
| 55 | + epoch_str = ( |
| 56 | + "Epoch %d. Train Loss: %f, Train Acc: %f, Valid Loss: %f, Valid Acc: %f, " |
| 57 | + % (epoch, train_loss / len(train_data), |
| 58 | + train_acc / len(train_data), valid_loss / len(valid_data), |
| 59 | + valid_acc / len(valid_data))) |
| 60 | + else: |
| 61 | + epoch_str = ("Epoch %d. Train Loss: %f, Train Acc: %f, " % |
| 62 | + (epoch, train_loss / len(train_data), |
| 63 | + train_acc / len(train_data))) |
| 64 | + prev_time = cur_time |
| 65 | + print(epoch_str + time_str) |
| 66 | + |
| 67 | + |
| 68 | +trans_train = transforms.Compose( |
| 69 | + [transforms.RandomResizedCrop(224), |
| 70 | + transforms.RandomHorizontalFlip(), |
| 71 | + transforms.ToTensor(), |
| 72 | + transforms.Normalize(mean=[0.485, 0.456, 0.406], |
| 73 | + std=[0.229, 0.224, 0.225])]) |
| 74 | + |
| 75 | +trans_valid = transforms.Compose( |
| 76 | + [transforms.Resize(256), |
| 77 | + transforms.CenterCrop(224), |
| 78 | + transforms.ToTensor(), |
| 79 | + transforms.Normalize(mean=[0.485, 0.456, 0.406], |
| 80 | + std=[0.229, 0.224, 0.225])]) |
| 81 | + |
| 82 | +trainset = torchvision.datasets.CIFAR10(root='../data', train=True, |
| 83 | + download=False, transform=trans_train) |
| 84 | +trainloader = torch.utils.data.DataLoader(trainset, batch_size=64, |
| 85 | + shuffle=True, num_workers=2) |
| 86 | + |
| 87 | +testset = torchvision.datasets.CIFAR10(root='../data', train=False, |
| 88 | + download=False, transform=trans_valid) |
| 89 | +testloader = torch.utils.data.DataLoader(testset, batch_size=64, |
| 90 | + shuffle=False, num_workers=2) |
| 91 | + |
| 92 | +classes = ('plane', 'car', 'bird', 'cat', |
| 93 | + 'deer', 'dog', 'frog', 'horse', 'ship', 'truck') |
| 94 | + |
| 95 | +# # 随机获取部分训练数据 |
| 96 | +# dataiter = iter(trainloader) |
| 97 | +# images, labels = dataiter.next() |
| 98 | + |
| 99 | +# 使用预训练的模型 |
| 100 | +net = models.resnet18(pretrained=True) |
| 101 | + |
| 102 | +# Freeze model weights |
| 103 | +for param in net.parameters(): |
| 104 | + param.requires_grad = False |
| 105 | + |
| 106 | +# 将最后的全连接层改成十分类 |
| 107 | +device = torch.device("cuda:1" if torch.cuda.is_available() else "cpu") |
| 108 | +net.fc = nn.Linear(512, 10) |
| 109 | +#net = torch.nn.DataParallel(net) |
| 110 | + |
| 111 | +# 查看总参数及训练参数 |
| 112 | +total_params = sum(p.numel() for p in net.parameters()) |
| 113 | +print('总参数个数:{}'.format(total_params)) |
| 114 | +total_trainable_params = sum(p.numel() for p in net.parameters() if p.requires_grad) |
| 115 | +print('需训练参数个数:{}'.format(total_trainable_params)) |
| 116 | + |
| 117 | +net=net.to(device) |
| 118 | + |
| 119 | + |
| 120 | +criterion = nn.CrossEntropyLoss() |
| 121 | +#只需要优化最后一层参数 |
| 122 | +optimizer = torch.optim.SGD(net.fc.parameters(), lr=1e-3, weight_decay=1e-3,momentum=0.9) |
| 123 | + |
| 124 | + |
| 125 | +train(net, trainloader, testloader, 20, optimizer, criterion) |
0 commit comments