用Mnist数据测试VGG-16的准确性

电脑之前烧坏过，今年刚换的主板和散热器，看到cpu占用率100%不敢跑太多数据，所以选了1000个批次，每个批次取了4个图片，额应该是这样，反正是减少了数据集和批次大小。

最后准确率13%有点意外，因为还不是我自己训练的数据，是直接下载预训练的数据

 

import torch
import torchvision
import torchvision.transforms as transforms
import torch.nn as nn
from torchvision.models import vgg16

# 加载MNIST数据集并进行预处理
transform = transforms.Compose([
    transforms.Resize((224, 224)),  # 将图像调整为VGG-16所需的输入尺寸
    transforms.Grayscale(num_output_channels=3),  # 将图像转换为3通道的灰度图像
    transforms.ToTensor(),
    transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])

testset = torchvision.datasets.MNIST(root='./data', train=False, download=True, transform=transform)

subset_indices = range(1000)  # 选择前1000个示例作为子集
subset_sampler = torch.utils.data.SubsetRandomSampler(subset_indices)
testloader = torch.utils.data.DataLoader(testset, batch_size=4, sampler=subset_sampler, num_workers=0)

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# 加载预训练的VGG-16模型
model = vgg16(pretrained=True)

# 修改第一层卷积层的输入通道数
model.features[0] = nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1)

# 修改最后一层全连接层的输出维度
model.classifier[6] = nn.Linear(4096, 10)

model.to(device)
model.eval()  # 切换为评估模式

correct = 0
total = 0

with torch.no_grad():
    for data in testloader:
        images, labels = data[0].to(device), data[1].to(device)
        outputs = model(images)
        _, predicted = torch.max(outputs.data, 1)
        total += labels.size(0)
        correct += (predicted == labels).sum().item()

accuracy = 100 * correct / total
print('Accuracy of the VGG-16 model on the MNIST test images: %.2f %%' % accuracy)