【pytorch】transform的使用

一、transforms的用法

transforms 是数据预处理与增强的核心工具，主要用于将原始图像转换为模型可接受的格式，并通过随机变换丰富数据集以提高模型泛化能力。

导入方式：

python 复制代码

from torchvision import transforms

主要用法，按顺序

python 复制代码

transform_pipeline = transforms.Compose([
    transforms.Resize(256),          # 调整图像大小至256x256（保持宽高比）
    transforms.CenterCrop(224),      # 从中心裁剪224x224区域（常用预训练模型输入尺寸）
    transforms.RandomHorizontalFlip(p=0.5),  # 以50%概率水平翻转（数据增强）
    transforms.ToTensor(),           # 将PIL图像转换为Tensor（像素值缩放至[0,1]）
    transforms.Normalize(            # 标准化（使用ImageNet均值/方差）
        mean=[0.485, 0.456, 0.406],  # RGB通道均值
        std=[0.229, 0.224, 0.225]    # RGB通道标准差
    )
])

二、transform的使用

将PIL图像转换成Tensor类型

python 复制代码

from PIL import Image
from torchvision import transforms

img_path = r'data/train/ants_image/0013035.jpg'
img = Image.open(img_path)
tensor_trans = transforms.ToTensor()
tensor_img = tensor_trans(img)
print(tensor_img.shape)    #CHW

通过tensor()类型的数据生成tensorboard图

python 复制代码

from PIL import Image
from torch.utils.tensorboard import SummaryWriter
from torchvision import transforms

img_path = r'data/train/ants_image/0013035.jpg'
img = Image.open(img_path)
tensor_trans = transforms.ToTensor()
tensor_img = tensor_trans(img)
# print(tensor_img.shape)    #CHW
writer = SummaryWriter('logs')
writer.add_image('tensor_img', tensor_img, 0)
writer.close()

Normalize()归一化使用

python 复制代码

from PIL import Image
from torch.utils.tensorboard import SummaryWriter
from torchvision import transforms

img_path = r'data/train/ants_image/0013035.jpg'
img = Image.open(img_path)
tensor_trans = transforms.ToTensor()
tensor_img = tensor_trans(img)
# print(tensor_img.shape)    #CHW

writer = SummaryWriter('logs')
norm_trans = transforms.Normalize([0.485, 0.456, 0.406], [0.5, 0.5, 0.5])
norm_img = norm_trans(tensor_img)

writer.add_image('tensor_img', tensor_img, 0)
writer.add_image('norm_img', norm_img, 1)
writer.close()

归一化后的图片和未归一化的图片

Resize()调整大小的使用

python 复制代码

from PIL import Image
from torch.utils.tensorboard import SummaryWriter
from torchvision import transforms

img_path = r'data/train/ants_image/0013035.jpg'
img = Image.open(img_path)
tensor_trans = transforms.ToTensor()
tensor_img = tensor_trans(img)
# print(tensor_img.shape)    #CHW

writer = SummaryWriter('logs')
norm_trans = transforms.Normalize([0.485, 0.456, 0.406], [0.5, 0.5, 0.5])
norm_img = norm_trans(tensor_img)

# print(img.size)
resize_trans = transforms.Resize((256, 256))
resize_img = resize_trans(tensor_img)
writer.add_image('resize_img', resize_img, 0)
# print(resize_img.size)
#Compose用法
trans_resize_2 = transforms.Compose([transforms.Resize((512)), transforms.ToTensor()])
img_resize_2 = trans_resize_2(img)


writer.add_image('tensor_img', tensor_img, 0)
writer.add_image('norm_img', norm_img, 1)
writer.add_image('img_resize_2', img_resize_2, 2)
writer.close()