1、旧版
2、新版
3、新旧版对比
补充：使用numpy模拟torch.fft.fft拯救paddle
总结
在新旧版的torch中的傅里叶变换函数在定义和用法上存在不同，记录一下。
1、旧版fft = torch.rfft(input, 2, normalized=True, onesided=False)
# input 为输入的图片或者向量，dtype=torch.float32，size比如为[1,3,64,64]
# signal_ndim(int):The number of dimensions in each signal,can only be 1、2、3
# normalized(bool,optional):controls wheather to return normallized results. Default:False
# onesided(bool,optional):controls whether to return half of results to avoid redundancy.Default:True
上面例子中图像中 singal_ndim = 2 ,是因为输入图像是2维的。
1.7之后的版本中，如果要用 oneside output，则改用torch.fft.rfft();如果要用two-side output，则改用torch.fft.fft()
input= torch.arange(4)
fft = torch.rfft(input, 2, normalized=True, onesided=False)
2、新版
一维离散傅里叶变换
torch.fft.rfft(input,n=None,dim=-1,norm=None) --> Tensor
# input:Tensor
# n(int,optional):Output signal length. This determines the length of the
output signal.
# dim(int, optional): The dimension along which to take the one dimensional real IFFT.
# norm (str, optional): Normalization mode.
二维离散傅里叶变换
torch.fft.rfft2(input, s=None, dim=(-2, -1), norm=None, *, out=None) -> Tensor
input (Tensor): the input tensor
s (Tuple[int], optional): Signal size in the transformed dimensions.
dim (Tuple[int], optional): Dimensions to be transformed.
norm (str, optional): Normalization mode.
高维离散傅里叶变换
rfftn(input, s=None, dim=None, norm=None, *, out=None) -> Tensor
input (Tensor): the input tensor
s (Tuple[int], optional): Signal size in the transformed dimensions.
dim (Tuple[int], optional): Dimensions to be transformed.
norm (str, optional): Normalization mode. For the forward transform
3、新旧版对比
import torch
input = torch.rand(1,3,32,32)
# 旧版pytorch.rfft()函数
fft = torch.rfft(input, 2, normalized=True, onesided=False)
# 新版 pytorch.fft.rfft2()函数
output = torch.fft.fft2(input, dim=(-2, -1))
output = torch.stack((output.real, output_new.imag), -1)
ffted = torch.rfft(input, 1, onesided=False) to ffted = torch.view_as_real(torch.fft.fft(input, dim=1))
and
iffted = torch.irfft(time_step_as_inner, 1, onesided=False) to
iffted = torch.fft.irfft(torch.view_as_complex(time_step_as_inner), n=time_step_as_inner.shape[1], dim=1)
补充：使用numpy模拟torch.fft.fft拯救paddle
import numpy as np
import torch
import paddle
def paddle_fft(x,dim=-1):
if dim==-1:
return paddle.to_tensor(np.fft.fft(x.numpy()))
else:
shape= [i for i in range(len(x.shape))]
shape[dim],shape[-1]=shape[-1],shape[dim]
x=np.transpose(np.fft.fft(np.transpose(x.numpy(), shape)),shape)
return paddle.to_tensor(x)
if __name__ == '__main__':
data=paddle.to_tensor(np.array([[[1, 4, 3], [1, 2, 3]], [[1, 2, 3], [1, 2, 3]]]))
paddle_f_d=paddle_fft(paddle_fft(data,-1),-2)
torch_f_d =paddle_fft(torch.fft.fft(torch.Tensor(data.numpy()),dim=-1),-2)
print(paddle_f_d.numpy())
print(torch_f_d.numpy())
总结
到此这篇关于Python torch.fft.rfft()函数用法的文章就介绍到这了,更多相关torch.fft.rfft()函数用法内容请搜索易知道(ezd.cc)以前的文章或继续浏览下面的相关文章希望大家以后多多支持易知道(ezd.cc)！
