-
Hello,
I love DirectML but I'm experiencing slow depthwise convolutions (with group count equal to channel count) with DirectML 1.8.2 compared to regular convolutions, when in fact the depthwise co…
-
On CUDA, when the convolution batching rule uses group convolutions, this sometimes ends up being slower that we expect on older hardware. This is probably because PyTorch's group convolution calls th…
-
## 🚀 Feature
An FFT backend for `kornia.filters.filter2d`.
## Motivation
`kornia.filters.filter2d` is very slow for large kernels as it currently only performs convolutions in the spatial domai…
-
Your library is pretty cool, but looks like it was not updated for a long period of time.
At the same time, the version of libdnn in your Caffe fork seems to be more maintained and even got some ne…
romix updated
6 years ago
-
https://paperswithcode.com/method/depthwise-separable-convolution#:~:text=While%20standard%20convolution%20performs%20the,a%20linear%20combination%20of%20the
current setup
 a bit:
> That trick works because real space convolution is very memory and compute expensive: You have t…
-
Currently, the MDS layers of size `8, 12, 16, 24, 32, 64` are implemented by doing a convolution with an MDS vector (Meaning a vector whose associated circulant matrix is MDS) `v` of appropriate size.…
-
See [here](https://github.com/JuliaImages/ImageFiltering.jl/issues/52) for a description of the operations involved.
I expect that pooling will be relatively easy, convolutions are also straightfor…
-
### Describe the bug
A decent chunk of time in the Conformer model at training time is spent in the convolution module. Of that, a decent chunk is in the depthwise convolution, which sets `groups` to…
-
Current implementation of fused winograd convolution uses very limited subset of underlying kernel implementation.
Current limitations are:
- only 2x3 version
- no dilation support
- no grouped …