Current Server Time: 22:27 (Central Europe)

#apertus IRC Channel Logs

2021/01/13

Timezone: UTC


00:23
mumptai
left the channel
04:34
Bertl
off to bed now ... have a good one everyone!
04:34
Bertl
changed nick to: Bertl_zZ
06:50
BAndiT1983|away
changed nick to: BAndiT1983
06:54
illwieckz
left the channel
06:55
illwieckz
joined the channel
08:19
se6ast1an
good day
08:48
mumptai
joined the channel
11:01
BAndiT1983
changed nick to: BAndiT1983|away
11:04
BAndiT1983|away
changed nick to: BAndiT1983
11:42
mumptai
left the channel
11:55
berto_bxl
joined the channel
11:56
berto_bxl54
joined the channel
12:00
berto_bxl
left the channel
12:06
Bertl_zZ
changed nick to: Bertl
12:06
Bertl
morning folks!
12:09
BAndiT1983
hi
12:48
EmilJ
Hi all. I'll flash the xilinx-provided image and export the devicetree and kernel boot args. However since I'm not even getting into the u-boot shell I don't have a lot of faith that the issue starts only at kernel boot time
12:49
vup2
EmilJ: can you upload those somewhere?
12:49
EmilJ
I will
13:50
RexOrCine
left the channel
13:53
RexOrCine
joined the channel
14:19
eppisai
joined the channel
14:53
eppisai
left the channel
15:14
eppisai
joined the channel
15:31
eppisai
left the channel
16:06
EmilJ
vup2: https://gitlab.com/-/snippets/2060514
16:41
ZNC_
left the channel
16:42
ZNC
joined the channel
16:42
ZNC
changed nick to: Guest8416
16:43
BAndiT1983
left the channel
16:46
BAndiT1983
joined the channel
17:05
BAndiT1983
changed nick to: BAndiT1983|away
17:39
berto_bxl54
left the channel
18:24
BAndiT1983|away
changed nick to: BAndiT1983
18:42
Bertl
off for now ... bbl
18:43
Bertl
changed nick to: Bertl_oO
18:49
RexOrCine
left the channel
18:49
RexOrCine
joined the channel
19:13
anuejn
Bertl_oO: do you know how different the gains are for the beta greens?
19:15
anuejn
and se6ast1an do you have a link of dngs from the micro for me?
19:15
anuejn
I cannot find any on my computer
19:15
anuejn
looks innocent
19:18
se6ast1an
you mean the files you recorded ?
19:20
anuejn
yup
19:20
se6ast1an
uploading to https://cloud.apertus.org/index.php/s/KjJ8NLqpnPA5xjQ
19:20
anuejn
I only need one
19:21
anuejn
thanks :)
19:21
anuejn
how is software able to debayer the dngs from the beta correctly?
19:21
anuejn
it seems the green pixels in the even / odd rows have different gains
19:23
se6ast1an
do you mean which software?
19:23
se6ast1an
or how the software does it?
19:26
anuejn
how software generally does this
19:26
anuejn
eg. darktable
19:28
se6ast1an
there are several standard algorithms: bilinear, AHD, VNG, etc.
19:28
se6ast1an
ufraw is quite handy for choosing them individually
19:28
se6ast1an
the different green gains can lead to maze/grid patterns
19:28
se6ast1an
see examples here: https://sites.google.com/site/cornerfix/using-cornerfix/maze-patterns-1
19:29
mumptai
joined the channel
19:29
se6ast1an
it can easily be corrected by averaging the two green values in one 2x2 pixel block for example
19:30
se6ast1an
more sophisticated methods surely exist as well
19:30
anuejn
but that gives bad resolution loss?
19:30
anuejn
isnt the different gain a theoretically known factor that could be corrected for?
19:31
se6ast1an
yes
19:31
se6ast1an
the "VNG4" algorithm is known to be resistant to green-crosstalk
19:31
se6ast1an
its one option in ufraw
19:32
se6ast1an
the FPN and PRNU calibration should cancel it out
19:32
se6ast1an
the green differences I mean
19:33
anuejn
ah I guess I am just dumb and interpreted the dng file I had in the wrong way :(
19:33
anuejn
I was thinking that other channels are green than what they really are
19:34
se6ast1an
ah :)
19:43
anuejn
now it looks good
19:43
anuejn
and there is no "gain difference"
19:43
anuejn
sorry for the hassle
19:46
se6ast1an
great
19:46
se6ast1an
images to share?
19:47
anuejn
I am currently not writing any dngs
19:47
anuejn
but I can try :)
19:53
se6ast1an
you are working on the recorder software?
19:56
anuejn
no on the compression
19:57
anuejn
and I tried if it brings any improvements if we interpret the two green planes as one big
19:57
anuejn
spoiler: it doesnt
19:57
anuejn
(at least thats what I concluded)
20:09
se6ast1an
interesting
20:19
se6ast1an
There is the classic approach of chroma subsampling, mode like YCrBb : 4:2:2
20:20
se6ast1an
but I assume that does not translate directly to a raw colorspace
20:25
se6ast1an
should I explain?
20:26
anuejn
I do know how chroma subsampling works
20:26
anuejn
and red does something similiar in their patent with the raw data
20:26
anuejn
but I have to say that I do not really understand why the approach they are describing is good for non black and white images
20:27
se6ast1an
interesting!
20:30
se6ast1an
are you still encoding cineform or is this already outside their standard?
20:31
anuejn
the encoding scheme is from the cineform standart (with a few minor differences)
20:31
anuejn
but the data layout is different
20:32
anuejn
also we are currently skipping some part of the encoding process (some non linear coding of hf coefficients)
20:33
se6ast1an
but do you plan to produce files in the end that can be decoded by the cineform raw software?
20:33
anuejn
probably not
20:33
anuejn
since there seems to be not a lot of benefit and cineform is a huge and really dirty codebase
20:34
anuejn
building something compatible with that seems like not a thing I want to do
20:34
se6ast1an
well it means the footage will not be editable with any established software then...
20:35
anuejn
can anyone really open cineform?
20:35
anuejn
but I really liked the idea of having a frameserver
20:36
anuejn
and then doing something like cinemaDNG
20:36
se6ast1an
let me research what software can directly open cineform raw currently
20:56
BAndiT1983
haven't we researched MLV in 2018 while GSoC?
20:57
BAndiT1983
nice, remembered GEDI (green edge directed interpolation) and our link is first one, even before the PDF we used for it: https://wiki.apertus.org/index.php/OpenCine.Green_Edge_Directed_Interpolation
21:08
se6ast1an
ok, research complete
21:08
se6ast1an
gopro has killed cineform raw
21:08
se6ast1an
I couldnt get it to open anywhere
21:09
se6ast1an
resolve works with the non raw cineform files out of the box
21:09
se6ast1an
but not the raw one
21:09
se6ast1an
adobe, vlc cant import it either
21:09
se6ast1an
after installing a legacy gopro studio software
21:10
se6ast1an
I can ope the cineform raw files in the adobe suite but it crashes immediately then
21:10
se6ast1an
the gopro studio software cannot open cineform raw files...
21:10
se6ast1an
so the frame server sounds like a good approach :D
21:12
anuejn
:D
21:12
anuejn
fun
21:12
anuejn
after all raw files dont seem like gopros main market segment
21:13
anuejn
tho native cineform raw support would be nice because it allows lower res preview debayering for free
21:14
anuejn
maybe one day we have to add support for our file format in some software ;)
21:19
se6ast1an
well the frameserver would work around exactly that no?
21:22
anuejn
hm... yeah but that wouldnt be as integrated
21:22
anuejn
for example one could not just transparently use different resolutions for playback and rendering
21:25
BAndiT1983
some programs can reload data when they notice changes, so theoretically you can change the header or wherever the resolution is stored, when the user selects different options in the frameserver
21:26
BAndiT1983
or the user also has to reload the clips in the editor, e.g. blender VSE
21:29
anuejn
I tried the blender vse but somehow it was really slow on my machiene
21:29
anuejn
I wasnt able to do fhd playback with color correction at 30fps
21:39
BAndiT1983
seems to be a general problem there -> https://github.com/oormicreations/VSRender
21:40
BAndiT1983
had not many problems in 2012 with FHD and bluescreen, but maybe my memory betrays me
21:41
BAndiT1983
VSE uses multicore, but is also not optimized, at least this is what a quick google search telling me and that's how i've landed on github as someone suggested that addon
21:43
BAndiT1983
for frameserver (gsoc2019) i was considering opencl, or similar, to process the frames faster, plus downscaling like after effects does it, user can select /2, /4 or /8 resolution to lower CPU usage, additionally quick debayering would help to see the preview, but when high-res wide color is required, then frameserver could be set to quality mode for final processing
21:43
BAndiT1983
were just ideas, as the reality depends also on video software and capabilities
22:03
anuejn
se6ast1an: https://files.niemo.de/axiomlabs-roundtripped.dng
22:03
anuejn
vs https://files.niemo.de/axiomlabs-roundtripped.dng
22:04
anuejn
roundtripped is compressed circa 1:4 - 1.5
22:04
BAndiT1983
first one shows additional infos, like resolution and color info, in rawtherapee, any difference in file format?
22:05
anuejn
nope this exactly the same dng with exactly the same metadata
22:05
anuejn
(different one than the original frame but somehow dng writing / reading in python is pita because there are no god libs)
22:05
BAndiT1983
interesting
22:06
BAndiT1983
just a moment
22:07
BAndiT1983
the links are the same
22:07
BAndiT1983
but why the hell is rawtherapee showing more info for one of the files? have to do binary diff
22:07
anuejn
oh I see I pasted the wrong link
22:08
anuejn
second should be https://files.niemo.de/axiomlabs-original.dng
22:10
BAndiT1983
maybe a dumb question, as i wasn't following latest discussion very closely, but what is the difference? size is around 12MB, also from the quick glance while zooming in not that different
22:11
anuejn
the images were compressed to a compressed intermediate format and than converted back to dng
22:11
anuejn
so the size is exactly the same
22:11
anuejn
this is a test to show how visually lossy the compression scheme is
22:12
anuejn
more files in https://files.niemo.de/compare_dng/
22:12
anuejn
with compression ratios in https://files.niemo.de/compare_dng/ratios.txt
22:13
anuejn
this is for some random quantization values which are not optimized for any purpose
22:16
BAndiT1983
thanks for the explanation, will check how to check diff visually
22:17
BAndiT1983
ehm, bad english indeed
22:17
anuejn
hm... generally the images have to much high frequency parts after compression (looking at the eye of the portrait with the glasses)
22:18
anuejn
this might be a thing where we generally want to try remove hf when in doubt as it looks better ;)
22:18
anuejn
BAndiT1983: you can load the folder in rawTherapee. it will keep your current zoom while switching betwen images
22:18
BAndiT1983
haven't found any useful info recently, when Bertl mentioned removing noise and adding it after processing
22:19
BAndiT1983
ah, right, this would also help, google also suggested digikam as it syncs panning and zooming
22:20
anuejn
also AMaZE seems to interact badly with the compression
22:21
anuejn
with VNG4 I see a lot less artifacts
22:21
BAndiT1983
is it some lib which provides them?
22:21
Bertl_oO
hmm, different gains for the geens sounds odd ...
22:22
BAndiT1983
confirmed, if you select the images and choose "place on light table", then you can drag and drop into left and right pane, then zoom and pan, which is in sync
22:24
anuejn
Bertl_oO: it was just me being stupid decoding some dng, nvm
22:24
danieel
left the channel
22:25
Bertl_oO
anuejn: ah, okay, happens ...
22:25
anuejn
BAndiT1983: you can select them in a dropdown menu in RawTherapee
22:25
danieel
joined the channel
22:26
BAndiT1983
ah, my bad, forgot that DNG are not containing the processed data
22:29
se6ast1an
very nice!
22:30
se6ast1an
heading off for today
22:30
se6ast1an
good night
22:31
BAndiT1983
good night
22:32
BAndiT1983
indeed, AMaZE shows some very pronounced maze-like structures (probably where the name comes from), also AMaZE+VNG4 does not improve much there, but VNG4 itself is not that bad
22:33
BAndiT1983
hm, the structures do also appear with other methods, so probably comes from the compression
22:34
anuejn
yup atleast the debayering doesnt interact to well with it
22:34
BAndiT1983
hm, looked at leaf_original and i see same structures there
22:39
anuejn
yup those images are not noise free
22:39
BAndiT1983
looked also at portrait, question would be if the artifacts are noticeable when used as a whole image and not zoomed in
22:39
anuejn
and the debayering does not like that
22:39
BAndiT1983
some years ago, a camera man told us, that they usually want some more realistic look and it's not razor-sharp usually
22:40
BAndiT1983
the one who developed SHOODAK algo
22:40
BAndiT1983
https://www.apertus.org/what-is-debayering-article-october-2015
22:40
BAndiT1983
https://wiki.apertus.org/index.php/Shoodak
22:40
BAndiT1983
off for today, good night
22:40
BAndiT1983
changed nick to: BAndiT1983|away
22:41
anuejn
that basically falls into the category "add noise in post" ;)
22:41
anuejn
but definitely a valid approach
22:47
mumptai
left the channel