-
It would be great if we had a direct convolution kernel, which would probably be faster for small convolutions.
-
@liliwilson can u pls put the improvement suggestions here thx
-
-
If use_bias is set to True with ComplexConv2D (i.e. the default) the required time to train will be 2-3 times longer than if no bias is used. With keras real-valued Conv2D the difference due to bias i…
-
### Is your feature request related to a problem? Please describe.
Currently, the `scipy.signal.convolve2d` function does not support a boolean `dtype` for the `in2` input array. This limitation re…
-
Hi:
I am trying to quantize a torchvision model (it is a slightly modified version of torchvision RetinaNet model), but when I apply direct quantization I see a large accuracy loss between the origin…
-
Just tested it in IPython
```
import torch as t
conv2d = t.nn.Conv2d(32,32,3,1,1).cuda()
conv2d_depthwise = t.nn.Conv2d(32,32,3,1,1,groups=32).cuda()
inp = t.randn(2,32,512,512).cuda()
# w…
-
When running deepseg_sc using the default parameters, the centerline detection seems to take a fairly large amount of time to compute (about 6 min), while it is usually fast, especially with the SVM a…
-
Hi,
I'd like to use your library with models that are trained on 3d images. How hard do you think would it be for me to add support for 3d convolutions?
best regards
-
Hi, I'm bit confused about current implementations of the repo and implementations used/discussed in related papers. I'll just state what I think is true. Please correct me if I'm wrong.
* Flashcon…