fkodom fft-conv-pytorch issues

fkodom / fft-conv-pytorch

Implementation of 1D, 2D, and 3D FFT convolutions in PyTorch. Much faster than direct convolutions for large kernel sizes.

MIT License

478 stars 58 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

Speed of depth-wise convolution

#25 lim1011 opened 3 months ago
0
How to achieve overlap and add/save

#24 jelly114514 opened 5 months ago
0
License

#23 hmaarrfk closed 1 year ago
8
Using fft-conv hurts convergence

#22 OasisArtisan closed 1 year ago
2
Complex value support?

#21 StephenHogg opened 2 years ago
0
CUDA out of memory with complex_matmul

#20 aminaab96 opened 2 years ago
5
FFTConvTranspose

#19 tolusophy opened 2 years ago
0
Add padding=same, support half-precision input

#18 papkov closed 1 year ago
3
adaptively moves offset to the right device so that gpu can be used

#17 alexhagen closed 2 years ago
3
Bug Fix and Torch Compatibility

#16 fkodom closed 2 years ago
0
Add Benchmark Plots

#15 fkodom closed 2 years ago
0
Cleanup Unit Tests

#14 fkodom closed 2 years ago
0
Perform dilation with Kronecker product

#13 aretor closed 2 years ago
2
Add dilation parameter and tests

#12 aretor closed 2 years ago
7
feat: remove einsum for efficiency

#11 yoyolicoris closed 2 years ago
1
Made it as python package

#10 yoyolicoris closed 3 years ago
1
in_channels must be divisible by groups

#9 yoyolicoris opened 3 years ago
0
Frequency domain sub-sampling for strided convolution

#8 yoyolicoris closed 2 years ago
4
add plots for better benchmark visualisation

#7 antonfrancois closed 2 years ago
1
bug

#6 williamlzw closed 3 years ago
4
Autograd for complex matrix multiplication in Pytorch ?

#5 RobinhoodKi closed 3 years ago
3
Propagation of error becomes large very fast

#4 dwromero closed 3 years ago
1
Depth-wise separable convolution?

#3 vaesl closed 3 years ago
11
Stride

#2 fshamsafar closed 3 years ago
1
can't work on GPU?

#1 libonwpu closed 3 years ago
1