transformers 库进行图像分割的例子

以下是使用 transformers 库进行图像分割的例子：

1. 安装必要的库和模块：

!pip install transformers
!pip install torch torchvision

2. 导入必要的库和模块：

import torch
import torchvision
import matplotlib.pyplot as plt
from transformers import ViTFeatureExtractor, ViTForImageSegmentation

3. 加载数据集：

transform = torchvision.transforms.Compose([
    torchvision.transforms.Resize((224, 224)),
    torchvision.transforms.ToTensor(),
])

train_dataset = torchvision.datasets.CocoDetection(
    root='./data/train2017',
    annFile='./data/annotations/instances_train2017.json',
    transform=transform
)

test_dataset = torchvision.datasets.CocoDetection(
    root='./data/val2017',
    annFile='./data/annotations/instances_val2017.json',
    transform=transform
)

4. 加载模型和特征提取器：

feature_extractor = ViTFeatureExtractor.from_pretrained('google/vit-base-patch16-224')
model = ViTForImageSegmentation.from_pretrained('google/vit-base-patch16-224')

5. 定义训练函数：

def train(model, train_dataloader, optimizer, criterion, device):
    model.train()
    train_loss = 0
    for i, (inputs, targets) in enumerate(train_dataloader):
        inputs = inputs.to(device)
        targets = targets.to(device)
        optimizer.zero_grad()
        outputs = model(inputs)
        loss = criterion(outputs, targets)
        loss.backward()
        optimizer.step()
        train_loss += loss.item()
    return train_loss / len(train_dataloader)

6. 定义测试函数：

def test(model, test_dataloader, criterion, device):
    model.eval()
    test_loss = 0
    with torch.no_grad():
        for i, (inputs, targets) in enumerate(test_dataloader):
            inputs = inputs.to(device)
            targets = targets.to(device)
            outputs = model(inputs)
            loss = criterion(outputs, targets)
            test_loss += loss.item()
    return test_loss / len(test_dataloader)

7. 训练模型：

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model.to(device)

optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)
criterion = torch.nn.CrossEntropyLoss()

train_dataloader = torch.utils.data.DataLoader(train_dataset, batch_size=16, shuffle=True)
test_dataloader = torch.utils.data.DataLoader(test_dataset, batch_size=16, shuffle=False)

num_epochs = 10

train_losses = []
test_losses = []

for epoch in range(num_epochs):
    train_loss = train(model, train_dataloader, optimizer, criterion, device)
    test_loss = test(model, test_dataloader, criterion, device)
    train_losses.append(train_loss)
    test_losses.append(test_loss)
    print(f'Epoch {epoch+1}/{num_epochs}, Train Loss: {train_loss:.4f}, Test Loss: {test_loss:.4f}')

plt.plot(train_losses, label='Train Loss')
plt.plot(test_losses, label='Test Loss')
plt.legend()
plt.show()

8. 对单个图像进行预测：

image = torchvision.io.read_image('image.jpg')
inputs = transform(image).unsqueeze(0).to(device)
outputs = model(inputs)
segmentation_map = torch.argmax(outputs, dim=1).squeeze().cpu().numpy()
plt.imshow(segmentation_map)
plt.show()

以上是使用 transformers 库进行图像分割的例子，其中使用了 ViT 模型和 COCO 数据集。