fft-conv-pytorch
fft-conv-pytorch copied to clipboard
Add padding=same, support half-precision input
trafficstars
- Calculate padding size based on kernel for stride=1, dilation=1
- Convert half-precision signal and kernel to float, because rfftn does not support half at the moment
- Remove unnecessary signal copy in padding
Thanks for this contribution! Just FYI -- I'll probably get to look over this in the next few days. Super busy lately haha 👍
If an RGB image is convoluted, is it 2D or 3D? Can there be a demonstration about fast convolution of pictures
Is this going to be merged?