高级数据增强方法

• Cutout：在图像上随机遮挡一个矩形区域。
• Mixup：将两张图像按照一定比例进行线性混合，同时混合对应的标签。
• CutMix：将一张图像的矩形区域剪切并粘贴到另一张图像上，同时混合标签。
• Random Erasing：在图像中随机擦除一个区域。
• TransMix: TransMix 是一种用于增强 Vision Transformer (ViT) 模型的高级数据增强方法。它结合了 Mixup 和 CutMix 的思想，并应用于 Transformer 的 attention 机制上。

Mixup 和 CutMix

import numpy as np
import torch

def mixup_data(x, y, alpha=1.0):
    '''Returns mixed inputs, pairs of targets, and lambda'''
    if alpha > 0:
        lam = np.random.beta(alpha, alpha)
    else:
        lam = 1

    batch_size = x.size()[0]
    index = torch.randperm(batch_size).cuda()

    mixed_x = lam * x + (1 - lam) * x[index, :]
    y_a, y_b = y, y[index]
    return mixed_x, y_a, y_b, lam

def cutmix_data(x, y, alpha=1.0):
    '''Returns cutmix inputs, pairs of targets, and lambda'''
    if alpha > 0:
        lam = np.random.beta(alpha, alpha)
    else:
        lam = 1

    batch_size = x.size()[0]
    index = torch.randperm(batch_size).cuda()

    bbx1, bby1, bbx2, bby2 = rand_bbox(x.size(), lam)
    x[:, :, bbx1:bbx2, bby1:bby2] = x[index, :, bbx1:bbx2, bby1:bby2]

    y_a, y_b = y, y[index]
    return x, y_a, y_b, lam

def rand_bbox(size, lam):
    '''Generate random bounding box'''
    W = size[2]
    H = size[3]
    cut_rat = np.sqrt(1. - lam)
    cut_w = np.int(W * cut_rat)
    cut_h = np.int(H * cut_rat)

    # uniform
    cx = np.random.randint(W)
    cy = np.random.randint(H)

    bbx1 = np.clip(cx - cut_w // 2, 0, W)
    bby1 = np.clip(cy - cut_h // 2, 0, H)
    bbx2 = np.clip(cx + cut_w // 2, 0, W)
    bby2 = np.clip(cy + cut_h // 2, 0, H)

    return bbx1, bby1, bbx2, bby2

应用 Mixup 或 CutMix 进行训练

for i, data in enumerate(trainloader, 0):
    inputs, labels = data
    inputs, labels = inputs.to(device), labels.to(device)

    # 使用Mixup
    inputs, targets_a, targets_b, lam = mixup_data(inputs, labels, alpha=1.0)
    # 或者使用CutMix
    # inputs, targets_a, targets_b, lam = cutmix_data(inputs, labels, alpha=1.0)

    inputs, targets_a, targets_b = map(torch.autograd.Variable, (inputs, targets_a, targets_b))

    optimizer.zero_grad()
    outputs = model(inputs)
    loss = mixup_criterion(criterion, outputs, targets_a, targets_b, lam)
    # 如果使用CutMix，使用相同的损失函数计算方法
    loss.backward()
    optimizer.step()

TransMix

导入必要的库

import torch
import torch.nn as nn
import torch.optim as optim
import torchvision
import torchvision.transforms as transforms
from vit_pytorch import ViT
import numpy as np

# 定义TransMix增强方法
def transmix_data(x, y, alpha=1.0):
    '''Returns transmix inputs, pairs of targets, and lambda'''
    if alpha > 0:
        lam = np.random.beta(alpha, alpha)
    else:
        lam = 1

    batch_size = x.size()[0]
    index = torch.randperm(batch_size).cuda()

    mixed_x = lam * x + (1 - lam) * x[index, :]
    y_a, y_b = y, y[index]

    # Generate random attention map
    attention_map = torch.rand_like(mixed_x)
    mixed_x = lam * x * attention_map + (1 - lam) * x[index, :] * (1 - attention_map)

    return mixed_x, y_a, y_b, lam

def transmix_criterion(criterion, pred, y_a, y_b, lam):
    return lam * criterion(pred, y_a) + (1 - lam) * criterion(pred, y_b)

数据准备

transform = transforms.Compose([
    transforms.RandomResizedCrop(224),
    transforms.RandomHorizontalFlip(),
    transforms.RandomRotation(15),
    transforms.ColorJitter(brightness=0.4, contrast=0.4, saturation=0.4, hue=0.1),
    transforms.ToTensor(),
    transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
])

trainset = torchvision.datasets.CIFAR10(root='./data', train=True, download=True, transform=transform)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=32, shuffle=True, num_workers=2)

testset = torchvision.datasets.CIFAR10(root='./data', train=False, download=True, transform=transform)
testloader = torch.utils.data.DataLoader(testset, batch_size=32, shuffle=False, num_workers=2)

训练和验证

criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=0.001)

def train(epoch):
    model.train()
    running_loss = 0.0
    for i, data in enumerate(trainloader, 0):
        inputs, labels = data
        inputs, labels = inputs.to(device), labels.to(device)

        inputs, targets_a, targets_b, lam = transmix_data(inputs, labels, alpha=1.0)
        inputs, targets_a, targets_b = map(torch.autograd.Variable, (inputs, targets_a, targets_b))

        optimizer.zero_grad()
        outputs = model(inputs)
        loss = transmix_criterion(criterion, outputs, targets_a, targets_b, lam)
        loss.backward()
        optimizer.step()

        running_loss += loss.item()
        if i % 100 == 99:    # 每100个batch打印一次
            print('[%d, %5d] loss: %.3f' % (epoch + 1, i + 1, running_loss / 100))
            running_loss = 0.0

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

    print('Accuracy of the network on the 10000 test images: %d %%' % (100 * correct / total))

for epoch in range(10):  # 训练10个epoch
    train(epoch)
    test()

这段代码展示了如何使用 TransMix 进行数据增强，并将其应用于 Vision Transformer 模型的训练过程中。TransMix 结合了 Mixup 和 CutMix 的思想，并通过随机生成的 attention map 来增强图像数据。这样可以提升模型的泛化能力，减少过拟合，并提高模型在不同数据分布上的表现。你可以根据具体的任务和数据集调整这些增强方法的参数和组合。