PyTorch学习笔记（三）——Transforms

对图像进行一些变换

 

1 结构

from torchvision import transforms

 常用的类：

• ToTensor类：把一个PIL的Image或者numpy数据类型的图片转换成 tensor 的数据类型

为什么需要tensor的数据类型：

它包装了神经网络所需要的一些参数，如_backward_hooks、_grad等(先转换成tensor数据类型，再进行训练)

• ToPILImage类：把一个图片转换成PIL Image
• Normalize类：归一化
• Resize类：尺寸变换
• CenterCrop类：中心裁剪

2 用法

2.1 ToTensor

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

# 绝对路径：E:\360MoveData\Users\BLACK\Desktop\学习资料\learn_pytorch\dataset\train\ants\0013035.jpg
# 相对路径：dataset/train/ants/0013035.jpg
img_path = "dataset/train/ants/0013035.jpg"
img = Image.open(img_path)
# print(img)

tensor_trans = transforms.ToTensor()
tensor_img = tensor_trans(img)
# print(tensor_img)

writer = SummaryWriter("logs")
writer.add_image("Tensor_img", tensor_img)
writer.close()

2.2 归一化（Normalize）

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

writer = SummaryWriter("logs")
img = Image.open("dataset/train/ants/0013035.jpg")
# print(img)

# ToTensor
trans_totensor = transforms.ToTensor()
img_tensor =trans_totensor(img)
writer.add_image("ToTensor",img_tensor)

# Normalize
trans_norm = transforms.Normalize([0.5,0.5,0.5],[0.5,0.5,0.5]) # 参数是均值和标准差，因为图片是RGB三信道，故传入三个数
img_norm = trans_norm(img_tensor)
writer.add_image("Normalize",img_norm)

writer.close()

公式：output[channel] = (input[channel] - mean[channel]) / std[channel]

举例：(input-0.5)/0.5=2*input-1

          input [0,1]

          result [-1,1]

2.3 Resize

将输入转变到给定尺寸

输入：PIL Image

• 序列：(h,w)
• 整数：不改变高和宽的比例，最小的边将会匹配这个数（等比缩放）

返回值：PIL Image

# Resize
# print(img.size)
trans_resize = transforms.Resize((512,512))
# img (PIL) -> resize -> img_resize(PIL)
img_resize = trans_resize(img)
# img_resize(PIL) -> ToTensor -> img_resize(tensor)
img_resize = trans_totensor(img_resize)
writer.add_image("Resize",img_resize,0)
# print(img_resize)

 2.4 Compose

• 把不同的transforms方法结合在一起
• 参数是一个数组，里面是不同的transforms方法（注意顺序，前一个的输出是下一个的输入）

举例：

Example:
        >>> transforms.Compose([
        >>>     transforms.CenterCrop(10),
        >>>     transforms.PILToTensor(),
        >>>     transforms.ConvertImageDtype(torch.float),
        >>> ])

# Compose
trans_resize_2 = transforms.Resize(512)
trans_compose = transforms.Compose([trans_resize_2,trans_totensor])
img_resize_2 = trans_compose(img)
writer.add_image("Resize",img_resize_2,1)

 2.5 RandomCrop

随机裁剪

输入：PIL Image

# RandomCrop
trans_random = transforms.RandomCrop(512)
# trans_random = transforms.RandomCrop((200,300))
trans_compose_2 = transforms.Compose([trans_random,trans_totensor])
# 裁剪10个
for i in range(10):
    img_crop = trans_compose_2(img)
    writer.add_image("RandomCrop",img_crop,i)