Current Server Time: 14:05 (Central Europe)

#apertus IRC Channel Logs

2018/07/02

Timezone: UTC


01:15
rton
left the channel
01:43
jucar
joined the channel
02:08
jucar
left the channel
05:00
Lordvan_
changed nick to: LordVan
07:36
se6astian|away
changed nick to: se6astian
07:58
se6astian
changed nick to: se6astian|away
08:14
sebix
joined the channel
08:14
sebix
left the channel
08:14
sebix
joined the channel
08:48
se6astian|away
changed nick to: se6astian
09:18
BogdanXOR
joined the channel
09:33
rton
joined the channel
09:47
RexOrCine|away
changed nick to: RexOrCine
10:23
BogdanXOR
left the channel
10:37
BogdanXOR
joined the channel
11:06
TofuLynx
joined the channel
11:07
TofuLynx
Good Morning!
11:08
TofuLynx
se6astian: Thanks for the photos! However the photos are glitchy, even in darktable. I converted using raw2dng. Did I miss any step?
11:17
TofuLynx
Nevermind, Fixed :)
11:17
TofuLynx
I had to use --swap-line
11:48
BogdanXOR
darktable vs rawtherapee - what are your thoughts on this? I have just some (I'd say limited) rawtherapee experience, so I'd like to take the opportunity to get more insights into these (or any other open source & free alternatives) as to decide on a path to follow
11:48
BogdanXOR
so what do you prefer / recommend, which is the better bet?
11:49
BogdanXOR
you = everyone here, feel free to share input
11:58
se6astian
both are good :)
11:58
se6astian
TofuLynx: great to hear
12:31
TofuLynx
BogdanXOR: never used rawtherapee, but darktable is good
13:10
comradekingu
Used both, darktable for me
13:11
comradekingu
Depends how many controls you want
13:17
BogdanXOR
could you elaborate on that (amount of controls)?
13:19
comradekingu
from what i remember rawtherapee didnt offer as much adjustment
13:22
BogdanXOR
I see. I should give darktable a spin then
13:22
BogdanXOR
any insights on the speed of rendering the image changes, between the two?
13:31
se6astian
I think rawtherapee is offering just as much control/development options by now
14:09
se6astian
added a header margin-top on wiki
14:09
se6astian
hopefully this will help structure pages a bit better
14:10
se6astian
please check and report back if you think its too much/too little
14:21
TofuLynx
https://wiki.apertus.org/index.php/AXIOM_Beta/Camera_Structure
14:21
TofuLynx
is this layout expected?
14:22
TofuLynx
the images seem a bit off
14:24
Bertl_zZ
changed nick to: Bertl
14:24
Bertl
morning folks!
14:24
TofuLynx
Hello Bertl :)
14:35
LordVan
left the channel
14:39
TofuLynx
Unrelated to apertus, anyone here is expert in network systems, namely Mobile Networks?
14:49
ansari
joined the channel
14:49
ansari
left the channel
14:55
BogdanXOR
left the channel
14:56
BogdanXOR
joined the channel
14:56
sebix
left the channel
14:58
BAndiT1983|away
changed nick to: BAndiT1983
15:32
ymc98_2
joined the channel
15:35
ymc98_1
left the channel
15:45
TofuLynx
BAndiT1983: Are you there?
15:47
BAndiT1983
hi TofuLynx
15:47
TofuLynx
Hey :)
15:47
TofuLynx
Remember the thing I said with the nested for loops?
15:48
BAndiT1983
yes
15:48
TofuLynx
The acceleration I made so far that is on github improved GEDI from 67ms to 9ms. Now, I used nested for loops to remove that if statements, and it is now just 5ms :)
15:49
BAndiT1983
very good
15:49
BAndiT1983
which optimizatin flag do you use?
15:49
BAndiT1983
*optimization
15:49
TofuLynx
none for now, just openMP
15:50
TofuLynx
when I use O3 or O2 it doesnt affect the speed at all, which is strange
15:50
BAndiT1983
in release mode?
15:50
TofuLynx
yeah
15:50
BAndiT1983
one of the next steps is to analyze whole processing pipeline times
15:51
TofuLynx
Okay
15:51
BAndiT1983
file loading -> conversion -> debayering etc.
15:51
TofuLynx
File loading and saving is memory bound, right?
15:51
BAndiT1983
to find more culprits, it's a preparation for dng folder loading
15:51
TofuLynx
so it cant really be accelerated, correct?
15:51
BAndiT1983
more disk bound, in my opinion
15:52
BAndiT1983
yep, meant by file loading the steps which are taken after it'S in the memory
15:52
TofuLynx
I was referring to disk memory, sorry xD
15:52
BAndiT1983
maybe memory mapping would be useful for it
15:52
TofuLynx
hmm such as extractor?
15:53
BAndiT1983
yes
15:53
TofuLynx
Understood
15:53
BAndiT1983
there are several loops here and there, but i don't want to use SIMD at the moment, for byte swapping and processing
15:54
BAndiT1983
which is possible option for the future, as processors have special instructions for such cases like ours
15:57
BAndiT1983
quick reboot
15:57
TofuLynx
But is it really neccessary such speed by SIMD at loading?
15:58
BAndiT1983
loading?
15:58
BAndiT1983
changed nick to: BAndiT1983|away
15:58
TofuLynx
Isn't the memory mapping stuff related to the loading pipeline?
15:59
BAndiT1983|away
changed nick to: BAndiT1983
15:59
BogdanXOR
left the channel
15:59
BAndiT1983
memory mapping is used for bigger files
16:00
BAndiT1983
simd was mentioned as possible improvement for extractor
16:00
TofuLynx
hmm okay
16:00
TofuLynx
I will use the improvements made in GEDI for bilinear and for shoodak
16:01
TofuLynx
then I will research for further improvements in the GEDI code and SHOODAK code
16:01
TofuLynx
and then I will document the new algorithms which will mark the end of what I proposed in my proposal. Then I can start working on that task
16:02
BAndiT1983
excellent
16:02
TofuLynx
Later I would like to do the video playing
16:02
BAndiT1983
alright, i will join this task maybe soon, have first to sort out my work etc.
16:04
TofuLynx
Great :)
16:06
BogdanXOR
joined the channel
16:06
TofuLynx
I will be right back
16:07
BAndiT1983
ok
16:07
BAndiT1983
changed nick to: BAndiT1983|away
16:09
BAndiT1983|away
changed nick to: BAndiT1983
16:16
BogdanXOR
left the channel
16:17
BAndiT1983
changed nick to: BAndiT1983|away
16:18
BAndiT1983|away
changed nick to: BAndiT1983
16:48
se6astian
meeting in 12 minutes
16:53
TofuLynx
Back
16:57
ArunM
helloooo
16:58
Bertl
hello ArunM!
16:59
TofuLynx
Hello ArunM :)
17:00
ArunM
Bertl are you available after the meeting?
17:00
Bertl
sure
17:00
se6astian
good evening everyone
17:00
se6astian
again please PM me now to be put in the sharing/reporting order
17:01
se6astian
nmdis199-: contacted me beforehand to be excused from todays meeting
17:01
se6astian
and first up is ArunM, please go ahead
17:02
ArunM
Hello everyone
17:02
ArunM
Currently programming the Generator part the design
17:02
ArunM
Here is the architecture
17:02
ArunM
https://docs.google.com/document/d/1Pv46Nvl1mE6RruxbesQJu-GaRoef1IV2VUDQu5oFSJo/edit?usp=sharing
17:02
ArunM
The architecture is yet to be discussed with BertL, already started programming because I wanted to check the possible patterns that can be generated
17:02
ArunM
Now having a better idea of the design and no problem if it have to be changed completely
17:03
ArunM
During simplification of old code i got stuck in changing one design for another and then changing next module.
17:03
ArunM
Finally relaxed with it and hopefully the documentation of old code will be submitted before next week evaluation.
17:03
ArunM
along with the generator module
17:03
ArunM
Here is the list of old patterns for the camera https://docs.google.com/document/d/1Ph_kRRllu43dGQx5Iw6vgpNmFYtquQEu-2h4mNFIS3c/edit?usp=sharing
17:04
ArunM
New patterns will be updated and discussed by tomorrow only.
17:04
ArunM
done reporting Captain :)
17:05
TofuLynx
How did you make the diagrams?
17:05
ArunM
draw.io
17:05
TofuLynx
oh nice :)
17:06
ArunM
also i'll be writting a detailed description b4 discussing
17:06
ArunM
the architecture
17:06
BAndiT1983
offline alternative to draw.io is yed
17:06
Bertl
sounds good!
17:06
se6astian
many thanks!
17:06
se6astian
ArunM: all done?
17:07
ArunM
yes
17:08
se6astian
right, TofuLynx you are next in line
17:08
TofuLynx
Okkey
17:08
TofuLynx
So my task for the last week was to accelerate Bilinear, GEDI and SHOODAK demosaicers
17:09
TofuLynx
I used openMP and my first approach
17:09
TofuLynx
which was giving each color channel to each thread
17:09
TofuLynx
made a significant improvement. 67ms to 27ms in case of GEDI debayer, on my PC
17:10
TofuLynx
however, BAndiT1983 quickly pointed that I could distribute the threads inside each loop
17:10
TofuLynx
I took this approach and it improved even more the speed. 67ms to 9ms in GEDI
17:11
TofuLynx
Yesterday and today I noticed that there was a condition in every loop that I thought it was slowing things down
17:11
TofuLynx
so I used nested for loops, which openMP also accelerates nicely
17:11
TofuLynx
At the end, it improved from 67ms to 5ms, which is great.
17:12
TofuLynx
In fps, it is equivalent to 15fps to 200fps
17:12
se6astian
very nice!
17:12
TofuLynx
Now, I will do a last look at the code
17:13
TofuLynx
and see where further improvements can be made
17:13
TofuLynx
Also, those benchmarks numbers can be "misleading"
17:13
TofuLynx
because, as openMP distributes the workload for the CPU threads
17:14
TofuLynx
my CPU has a fairly high number of threads, which is 12
17:14
TofuLynx
a more common computer only has four threads
17:14
TofuLynx
Nonetheless, it's a significant improvement
17:14
TofuLynx
That's all, se6astian :)
17:14
se6astian
great, thanks
17:15
se6astian
can you already estimate a 4096x3072 image processing time?
17:15
TofuLynx
the benchmarks were done with a 4096x3072 image
17:15
se6astian
ah, excellent
17:16
TofuLynx
so it's basically 200fps for a more than 4K video
17:16
TofuLynx
on the CPU, if we left the workload to the GPU, it would be even faster
17:16
TofuLynx
But that is not supported, for now
17:16
alexML
that sounds a lot faster than current hdmi4k / raw2dng stuff :D
17:16
TofuLynx
how fast is that?
17:17
alexML
you need to leave the computer overnight for a tiny clip :P
17:17
TofuLynx
wow xD
17:17
BAndiT1983
don't forget, that it'S just debayering, so numbers will shrink
17:18
TofuLynx
yeah, we have to have a fast memory pipeline
17:20
se6astian
anyone else who wants to report
17:20
se6astian
it would be your turn now
17:20
Bertl
ymc98_2: ping?
17:21
Bertl
rahul_: ping?
17:21
se6astian
I will use the time to report a bit myself: the next team talk video and article are almost ready for release
17:21
TofuLynx
supragya_: is this for logging?
17:21
BogdanXOR
joined the channel
17:21
se6astian
release should be in the next couple of days
17:21
TofuLynx
what are the topics of the team talk video?
17:22
se6astian
the move to the new office
17:22
se6astian
and you :)
17:22
se6astian
the gsoc students and projects
17:22
TofuLynx
Great! :)
17:22
se6astian
plus some other minor updates/news
17:23
se6astian
twitter posts about your recent progress:
17:23
se6astian
https://twitter.com/ApertusOSCinema/status/1011555886609944576
17:23
se6astian
910 impressions
17:23
se6astian
https://twitter.com/ApertusOSCinema/status/1013044232360210432
17:23
TofuLynx
what are impressions?
17:23
se6astian
4,729 impressions
17:24
se6astian
it means that it appeared on someones screen
17:24
TofuLynx
Oh cool!
17:24
BAndiT1983
and a comment for magiclantern forum :D
17:25
TofuLynx
the thing with canon 80d? xD
17:26
RexOrCine
Present
17:26
se6astian
ok that should conclude this weeks meeting then
17:26
se6astian
many thanks to all participants
17:26
se6astian
and rex in particular! :D
17:28
se6astian
TofuLynx: what do you think of the task of creating some screencaptures of OC in general and what you added so far?
17:34
TofuLynx
I can handle that :)
17:34
TofuLynx
For now, there's just one difference
17:34
TofuLynx
the ProcessingTest UI
17:34
TofuLynx
It was revamped by BAndiT1983
17:39
BogdanXOR
left the channel
17:59
se6astian
changed nick to: se6astian|away
18:00
RexOrCine
changed nick to: RexOrCine|away
18:16
TofuLynx
left the channel
18:20
BogdanXOR
joined the channel
18:55
se6astian|away
changed nick to: se6astian
19:09
BogdanXOR
left the channel
19:10
BogdanXOR
joined the channel
19:40
lexano
left the channel
19:41
se6astian
changed nick to: se6astian|away
19:43
BogdanXOR
left the channel
19:48
se6astian|away
changed nick to: se6astian
19:54
BogdanXOR
joined the channel
19:54
lexano
joined the channel
20:07
BogdanXOR
left the channel
20:41
BAndiT1983
changed nick to: BAndiT1983|away
20:41
BAndiT1983|away
changed nick to: BAndiT1983
20:56
se6astian
off to bed
20:56
se6astian
good night
20:56
se6astian
changed nick to: se6astian|away
21:12
BAndiT1983
changed nick to: BAndiT1983|away
22:02
BogdanXOR
joined the channel
23:15
BogdanXOR
left the channel
23:25
BogdanXOR
joined the channel
23:29
BogdanXOR
left the channel