Pytorch学习（四）Transforms用法学习

1. Transforms该如何使用?

Transforms用来对输入图片的格式及样式做一些改变，使用时需要注意图像的输入格式：

• PIL格式---->Image.open()
• ndarray------>cv.imread()
• 转换成Tensor格式-------->ToTensor()

from PIL import Image
from torchvision.transforms import transforms

img_path = "dataset/facades/train/a/cmp_b0005.png"   # 读取图片的路径
img = Image.open(img_path)
# 1. Transforms按照如下方式使用：
to_tensor = transforms.ToTensor()
tensor_img = to_tensor(img)                         # 转换成Tensor格式

print(tensor_img)

输出的Tensor值如下：

按照如下ToTensor类的定义，call()函数的调用，需要提供一个参数，参数类型是PIL.Image格式或者numpy.ndarray格式（使用opencv读取图片得到的格式就是ndarray）。

class ToTensor(object):
    """Convert a ``PIL Image`` or ``numpy.ndarray`` to tensor.

    Converts a PIL Image or numpy.ndarray (H x W x C) in the range
    [0, 255] to a torch.FloatTensor of shape (C x H x W) in the range [0.0, 1.0]
    if the PIL Image belongs to one of the modes (L, LA, P, I, F, RGB, YCbCr, RGBA, CMYK, 1)
    or if the numpy.ndarray has dtype = np.uint8

    In the other cases, tensors are returned without scaling.
    """

    def __call__(self, pic):
        """
        Args:
            pic (PIL Image or numpy.ndarray): Image to be converted to tensor.

        Returns:
            Tensor: Converted image.
        """
        return F.to_tensor(pic)

    def __repr__(self):
        return self.__class__.__name__ + '()'

2. 为什么要使用Tensor数据类型

Tensor数据类型有一些特殊的属性，如反向传播属性_backward_hooks，梯度属性_grad，requires_grad，Tensor对反向传播所需的参数进行包装，所以输入数据需要转换成tensor类型，进行训练。

3. Transforms常用的工具

如下代码依次测试了：

1. ToTensor()
2. Normalize()
3. Resize()
4. Compose()
5. RandomCrop()

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

writer = SummaryWriter("logs")
img = Image.open("oracles/menu.jpg")
print(img)

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

# 测试Normalize
trans_norm = transforms.Normalize([0.5,0.5,0.5], [0.5,0.5,0.5])
img_norm = trans_norm(img_tensor)
writer.add_image("Normalize", img_norm)

# 测试resize
print(img.size)        # 输出是(3456, 4608)
trans_resize = transforms.Resize((512,512))
img_resize = trans_resize(img)        # Resize()的输入是PIL Image
print(img_resize)     # 输出是(512,512)
img_resize = trans_totensor(img_resize)
writer.add_image("Resize", img_resize, 0)

# 测试Compose - resize -2
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)

# 测试RandomCrop随机裁剪
trans_random = transforms.RandomCrop(512)
trans_compose_2 = transforms.Compose([trans_random, trans_totensor])
for i in range(10):   # 0裁剪10次
    img_crop = trans_compose_2(img)
    writer.add_image("RandomCrop", img_crop, i)
writer.close()

使用TensorBoard展示结果：


两次Resize()的结果


随机裁剪了10次中最后一次裁剪结果：