in paifilter code: torch.fft.rfft(x, dim=1, norm='ortho') is on seq_len , but texfilter code torch.fft.rfft is calculated on feature_num ?

[wangty537]

[aikunyi]

right

[monsteryi0314]

According to the paper, shouldn't it be along the time dimension?

[aikunyi]

The filtering process is carried out along the time dimension

[zhujiayi18161778290]

So the original code is wrong? It should be x = torch.fft.rfft(x, dim=2, norm='ortho')？Only in this way can we filter along the time dimension.