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
Hi, I am looking at the correlation operation, which in the file xcorr.py.
I'm confused about the function xcorr_fast and xcorr_depthwise, that using F.con2d with group to speed up. The operation may indroduce some problems.
In detail, before each F.conv2d(), px will be reshaped to size of (1, batch * channel, w, h), pk is (batch, channel, w, h). Referring to the implemention of conv2d, for each channel of output, the result is the sum of the convolution of one kernel(channel, w, h) and all layers of the x which is totally batch * channel layers.
That's to say, for each batch of the output, it integrate the information of all other batches where I think every batch is independent.
Does this operation make sense?
Thank you!
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
-
Hi, I am looking at the correlation operation, which in the file
xcorr.py
.I'm confused about the function
xcorr_fast
andxcorr_depthwise
, that usingF.con2d
withgroup
to speed up. The operation may indroduce some problems.In detail, before each
F.conv2d()
,px
will be reshaped to size of(1, batch * channel, w, h)
,pk
is(batch, channel, w, h)
. Referring to the implemention ofconv2d
, for each channel of output, the result is the sum of the convolution of one kernel(channel, w, h) and all layers of the x which is totallybatch * channel
layers.That's to say, for each batch of the output, it integrate the information of all other batches where I think every batch is independent.
Does this operation make sense?
Thank you!
Beta Was this translation helpful? Give feedback.
All reactions