torchvision

1.torchvision简介

torchvision独立于pytorch，专门用来处理图像，通常用于计算机视觉领域。

1.1 torchvision中最常用的三个包

1.1.1 models

提供了很多常用的训练好的网络模型，我们可以直接加载并使用，如Alexnet、ResNet等。

1.1.2 datasets

提供了

• 一些常用的图片数据集，如MNIST、COCO等
• 加载自己的数据集的常用方法，目前只有DatasetFolder、ImageFolder、VisionDataset三个方法。

1.1.3 transforms

提供了一些常用的图像转换处理操作，主要针对Tensor或PIL Image进行操作。

2. torchvision 中的 transforms

2.1 transforms介绍

torchvision 库中的数据预处理工具主要体现在 torchvision.
transforms 模块。

它提供了丰富的函数和类来对图像数据进行各种形式的转换和预处理。这些预处理操作在深度学习中是至关重要的，因为它们可以增强模型的泛化能力，并且将不同大小和格式的原始图像数据转化为神经网络能够接受的标准输入。

2.2 ToTensor使用

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

# 绝对路径：C:\Users\awei0\Desktop\pytorch_to_learn\dataset\train\ants\0013035.jpg
# 相对路径：dataset/train/ants/0013035.jpg
# 1.transform 中ToTensor的使用
img_path = "dataset/train/ants/0013035.jpg"

#PIL类型图片

img = Image.open(img_path)
# print(type(img))

#PIL类型图片进行转换为Tensor
# 利用tensor_tans方法转化成tensor类型

trans_tensor = transforms.ToTensor()
tensor_img = trans_tensor(img)
# print(type(tensor_img))
print(tensor_img)

writer = SummaryWriter("logs1")
writer.add_image("Tensor_img", tensor_img)

writer.close()

命令行

tensorboard --logdir=logs1

2.3 Normalize使用

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_tensor = transforms.ToTensor()
img_tensor = trans_tensor(img)
writer.add_image("ToTensor", img_tensor)

# Normalize
# 因为img是RGB类型所以有三维
print(img_tensor[0][0][0])
# 给Normalize提供均值和标准差
# 按照``output[channel] = (input[channel] - mean[channel]) / std[channel]``进行计算
trans_norm = transforms.Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5])
# 进行归一化
img_norm = trans_norm(img_tensor)
print(img_norm[0][0][0])

writer.add_image("Normalize", img_norm)

writer.close()

Tensorboard 再次查看

可以看到上述经过归一化的图片

2.4 Resize

2.4.1 对元组

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

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

print(img)
print(img.size)

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

trans_resize = transforms.Resize((512, 512))
img_resize = trans_resize(img)
img_resize = trans_tensor(img_resize)
print(img_resize)


writer.add_image("Resize", img_resize, 0)
writer.close()

2.4.2 对单整型

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

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

print(img)
print(img.size)

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

# 1.元组
trans_resize = transforms.Resize((512, 512))
img_resize = trans_resize(img)
img_resize = trans_tensor(img_resize)
print(img_resize)


writer.add_image("Resize", img_resize, 0)
writer.close()

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

writer.close()

2.5RandomCrop随机裁剪

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

writer = SummaryWriter("logs1")
img_path = "dataset/train/ants/0013035.jpg"
img = Image.open(img_path)

print(img)
print(img.size)

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

#也可以投入元组()
trans_random = transforms.RandomCrop(512)
trans_compose = transforms.Compose([trans_random, trans_tensor])
for i in range(10):
    img_crop = trans_compose(img)
    writer.add_image("RandomCrop", img_crop, i)

writer.close()