You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
It would be very nice to replace the current dilation options
with a more intelligent one.
Currently the number of dilations can be specified,
and the dilations always increase.
There are multiple problems with this, but the two largest ones:
you cannot set the dilation
you cannot set the order
So instead of the current integer input it would be nice to accept two kind of inputs:
integer, which has the same meaning as currently
tuple, where the length of the tuple means the number of dilation steps,
and the numbers represent the pixel distance / dilation.
E.g. 1,2,4,1 means 4 blocks, with dilation of 1,2,4 and 1.
There are good arguments that a final refinement could be very helpful
to remove small scale noise. Also, the receptive field is better tunable
this way.
The implementation is relatively straightforward.
The text was updated successfully, but these errors were encountered:
Description
It would be very nice to replace the current dilation options
with a more intelligent one.
Currently the number of dilations can be specified,
and the dilations always increase.
There are multiple problems with this, but the two largest ones:
So instead of the current integer input it would be nice to accept two kind of inputs:
and the numbers represent the pixel distance / dilation.
E.g. 1,2,4,1 means 4 blocks, with dilation of 1,2,4 and 1.
There are good arguments that a final refinement could be very helpful
to remove small scale noise. Also, the receptive field is better tunable
this way.
The implementation is relatively straightforward.
The text was updated successfully, but these errors were encountered: