Alessio Sangalli wrote:
On Feb 6, 1:54 am, Alessio Sangallimano...@gmail.com wrote:
Right now my enfusion test case runs in about 0:47s down from the
original 1:36s
I did some profiling and I still get quite some FP usage (even if it
is much better than before)
Each sample counts as 0.01
On Feb 5, 10:44 pm, Alessio Sangalli mano...@gmail.com wrote:
I'd start by rewriting functors to use integer instead of float. For
making it even simpler I'd remove all functors used for determining
weight except for ExposureFunctor (by default it has highest weight
and for blending
On Feb 6, 1:54 am, Alessio Sangalli mano...@gmail.com wrote:
Right now my enfusion test case runs in about 0:47s down from the
original 1:36s
I did some profiling and I still get quite some FP usage (even if it
is much better than before)
Each sample counts as 0.01 seconds.
% cumulative
On 6 February 2011 10:54, Alessio Sangalli mano...@gmail.com wrote:
On Feb 5, 10:44 pm, Alessio Sangalli mano...@gmail.com wrote:
I'd start by rewriting functors to use integer instead of float. For
making it even simpler I'd remove all functors used for determining
weight except for
I will continue tomorrow with the analysys - bye, thank you!
as
I wish you luck. I can't be very helpful here, because it's quite some
time I hacked around enfuse but I'm glad that my suggestions has
helped you.
Lukas.
--
You received this message because you are subscribed to the Google
On 4 February 2011 20:50, Alessio Sangalli mano...@gmail.com wrote:
I was finally able to compile (statically!) all that is necessary to
run enfuse on an embedded ARM box. It's quite slow and a simple gprof
shows:
Each sample counts as 0.01 seconds.
% cumulative self self
On Feb 5, 1:01 am, Lukáš Jirkovský l.jirkov...@gmail.com wrote:
It's certainly doable. But I'm a bit surprised that it's using double
precision so much if your input data are stored as integers (take a
look at numerictraits.h [1]). I guess that the data you're getting
comes from functors
On Feb 5, 1:01 am, Lukáš Jirkovský l.jirkov...@gmail.com wrote:
It's certainly doable. But I'm a bit surprised that it's using double
precision so much if your input data are stored as integers (take a
Would you give me some more background, or advice what to debug for?
I am only now trying
On 5 February 2011 10:26, Alessio Sangalli mano...@gmail.com wrote:
On Feb 5, 1:01 am, Lukáš Jirkovský l.jirkov...@gmail.com wrote:
It's certainly doable. But I'm a bit surprised that it's using double
precision so much if your input data are stored as integers (take a
Would you give me
On Feb 5, 2:17 am, Lukáš Jirkovský l.jirkov...@gmail.com wrote:
I think I should point you to the Enfuse Algorithm page [1] which
contains description of how enfuse works with link to the paper in
PDF. All relevant code is in enfuse.h. Floating point arithmetics is
Thanks I've read the
I was finally able to compile (statically!) all that is necessary to
run enfuse on an embedded ARM box. It's quite slow and a simple gprof
shows:
Each sample counts as 0.01 seconds.
% cumulative self self total
time seconds secondscalls s/call s/call name
11 matches
Mail list logo