Pytorch实现上采用upsample和下采用downsample 简单调用函数即可实现，超简单的代码块调用

# 上采用函数，输入数据格式示例：tensor维度[3,300,300]，即3通道RGB，大小300×300，当然4通道图像也能做
def upsample(image_tensor, width, height, mode):
    # mode可用：最近邻插值"nearest"，双线性插值"bilinear"，双三次插值"bicubic"，如mode="nearest"
    image_upsample_tensor = torch.nn.functional.interpolate(image_tensor.unsqueeze_(0), size=[width, height], mode=mode)
    image_upsample_tensor.squeeze_(0)
    # 将数据归一到正常范围，尺寸改变过程可能数值范围溢出，此处浮点数据位[0,1]，整数数据为[0,255]
    image_upsample_tensor = image_upsample_tensor.clamp(0, 1)  
    return image_upsample_tensor


# 下采用函数，输入数据格式示例：tensor维度[3,300,300]，即3通道RGB，大小300×300，当然4通道图像也能做
def downsample(image_tensor, width, height):
    image_upsample_tensor = torch.nn.functional.interpolate(image_tensor.unsqueeze_(0), size=[width, height])
    image_upsample_tensor.squeeze_(0)
    # 将数据归一到正常范围，尺寸改变过程可能数值范围溢出，此处浮点数据位[0,1]，整数数据为[0,255]
    image_upsample_tensor = image_upsample_tensor.clamp(0, 1)
    return image_upsample_tensor

分享创造快乐，欢迎复制使用！