Closed rafaelorozco closed 1 year ago
Patch coverage: 95.83
% and project coverage change: +0.13
:tada:
Comparison is base (
df92d43
) 88.54% compared to head (ae9513b
) 88.68%.
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Do you have feedback about the report comment? Let us know in this issue.
@mloubout what do you think about merging this? I am mostly interested in the nx,nc,nbatch inputs because it looks like the (1,1,nx,nbatch) tensors is not natural for a lot of people. The dense nets could use help (want to give users easier options for swapping them out) but could be in a future PR.
That's fine just rebase it
Dense glow network Working for 1dim inputs, channel doesnt need to be divisible by 2. x = (nx,nc,batch) nc can be anything but if nc=1 then split_scales needs to be turned on to get extra channels to do the affine couplying layer
@mloubout from what I understand we need to feed in the nx dimension to define a dense layer right? I guess that was the magic of conv nets, that they only needed the channel dimension.