Edouard Oyallon
Edouard Oyallon
Awesome! This will be a great feature.
that's ok, spatial convolutions will be a future feature but I thinkdoing everything in fourier is more than fine
@rleonarduzzi where are we?
hi @rleonarduzzi
I'm in favor of the following (consistent behavior): If an input to a function is contiguous, the output should be contiguous. If it's not, bah, the output should just be...
Alright. Going to add a warning with low priority. I think it makes sens to add it in the "complex number PR" because basically relying on torch comple xnumbers allow...
This goes in the v3!
It's good to see the issue is compute for the first example of the gallery - for the scattering disk, I think the two pictures are similar yet the visualizations...
It's good
the disks are however quite different .. 👎 might be difficult to fix