自定义torchvision的transforms功能

在看FCN代码时，因为需要同时处理图像和模板标签，所以需要重写一系列的transforms的功能。

参照的是霹导的代码：

import torchvision.transforms.functional as F
class RandomResize(object):
    def __init__(self, min_size, max_size=None):
        self.min_size = min_size
        if max_size == None:
            max_size = min_size
        self.max_size = max_size

    def __call__(self, image, target):
        size = random.randint(self.min_size, self.max_size)
        image = F.resize(image, size)
        target = F.resize(target, size, interpolation=T.InterpolationMode.NEAREST)
        return image, target

我在浏览了一遍原代码后，试着自己写，写的过程发现，我用了transforms.Resize()，它的输入只有一个size变量，看起来不太对，我就返回看源码，才知道源码用的是transforms.functional。点进transforms.Resize()的源码发现，它的forward函数也调用的functional中的resize。

# transforms.Resize()源码
class Resize(torch.nn.Module):
    def __init__(self, size, interpolation=InterpolationMode.BILINEAR, max_size=None, antialias=None):
        super().__init__()
        _log_api_usage_once(self)
        if not isinstance(size, (int, Sequence)):
            raise TypeError(f"Size should be int or sequence. Got {type(size)}")
        if isinstance(size, Sequence) and len(size) not in (1, 2):
            raise ValueError("If size is a sequence, it should have 1 or 2 values")
        self.size = size
        self.max_size = max_size

        # Backward compatibility with integer value
        if isinstance(interpolation, int):
            warnings.warn(
                "Argument 'interpolation' of type int is deprecated since 0.13 and will be removed in 0.15. "
                "Please use InterpolationMode enum."
            )
            interpolation = _interpolation_modes_from_int(interpolation)

        self.interpolation = interpolation
        self.antialias = antialias

    def forward(self, img):
        return F.resize(img, self.size, self.interpolation, self.max_size, self.antialias)

结论，定义的时候用transforms中的函数，比如Compose时，如果要自己重写，需要调用transforms.functional中的函数，也就是transforms中对该函数的实现。