Current Server Time: 19:38 (Central Europe)

#apertus IRC Channel Logs

2018/05/20

Timezone: UTC


00:20
slikdigit
joined the channel
00:40
slikdigit_
joined the channel
00:41
slikdigit
left the channel
00:41
slikdigit_
changed nick to: slikdigit
01:20
rton
left the channel
02:05
slikdigit
left the channel
03:03
Bertl_oO
changed nick to: Bertl_zZ
05:11
BAndiT1983|away
changed nick to: BAndiT1983
09:23
slikdigit
joined the channel
09:39
rton
joined the channel
09:49
se6astian|away
changed nick to: se6astian
10:58
Bertl_zZ
changed nick to: Bertl
10:59
Bertl
morning folks!
11:37
Bertl
off for now ... bbl
11:37
Bertl
changed nick to: Bertl_oO
11:38
slikdigit
left the channel
12:11
TofuLynx
joined the channel
12:13
RexOrCine|away
changed nick to: RexOrCine
12:15
TofuLynx
Good Morning!
12:30
supragya
TofuLynx: hi!
12:38
TofuLynx
Hey supragya!
12:39
TofuLynx
I am going to lunch right now! Are you on this afternoon?
12:39
TofuLynx
(afternoon in europe)
12:39
supragya
*evening here :)
12:39
supragya
have meeting today... prepping for it
12:40
TofuLynx
uh nice
12:40
TofuLynx
meeting about what? anything important?
12:40
supragya
no, GSoC meet (our own), full meet on monday
12:41
ArunM
joined the channel
12:41
TofuLynx
Ah! :)
12:41
TofuLynx
Hope you have fun!
13:24
ymc98_1
left the channel
13:25
ymc98_1
joined the channel
13:32
BAndiT1983
hi TofuLynx, you had some questions regarding debayer?
13:51
ArunM
Meeting is tomorrow right?
13:51
ArunM
@10:30 ist
13:51
supragya
don't worry,
13:51
supragya
it's just for my proj (Raw video container format)
13:52
ArunM
okay :-)
13:58
TofuLynx
Oh hey BAndiT1983
13:59
TofuLynx
I didn't understand what you meant with this: "what about bilinear debayer for the blue channel, in our rggb example, which interpolates top, left and center sensel at once? then we would do everything in one pass, possibly multiple threads for performance. btw. the threads have to be in some distance from each other to avoid interference"
14:10
BAndiT1983
like it's described, you would interpolate the missing sensels, so the outer edge won'T be jagged
14:12
TofuLynx
but, how do you interpolate the blue channel for the border, while avoiding jagged lines?
14:24
supragya
hi BAndiT1983
14:24
BAndiT1983
hi supragya
14:25
BAndiT1983
TofuLynx: horizontally, in liner fashion, row 1 is meant (not 0)
14:25
supragya
BAndiT1983: I did in_memory emulation of the recorder to actually see how much perf we can get.. The very basic emulator was capable of having a throughput of about ~7500 frames per second
14:26
TofuLynx
Ok, so, for the borders, I will use linear interpolation, and for the row 1 and column 1 I will use bilinear intepolation (top-bottom and left-right), is that correct?
14:26
supragya
given a 4xSSD in RAID 0, we can expect somewhere around (x10 factor) ~750 frames of recording possible...
14:27
BAndiT1983
borders of blue, not the image border
14:28
BAndiT1983
supragya: don't forget that a PC and Zynq are totally different at thatz
14:28
BAndiT1983
*that
14:29
supragya
given that zync is capable of throughput anywhere below 750 frames per sec, we would be able to handle on recorder end
14:29
supragya
In the emulation, the AXIOM Beta was thought of as a black box... so no emulation was done to actually emulate how perf would depend on zync
14:30
TofuLynx
yep. But how can I do that in such a way that's universally applicable, with just offsets?
14:30
ArunM
left the channel
14:31
supragya
the stream given off by Beta in emulation was a simple fs (that's how it would look like when mounted through USB3, right?) and was programmed to give off as speedily the data as required.
14:32
supragya
g3gg0 in last meet asked to focus on the PC end... so I did that
14:36
BAndiT1983
TofuLynx: you should process it like in downscaler and such, as a block of 2x2 or possibly 3x3
14:42
TofuLynx
BAndiT1983: But in that case, How do I know where are the missing borders? For example, in RGGB, the red and green channel need the intepolation on the borders. But blue only needs on the top and left borders, and also in row 1 and column 1
14:44
BAndiT1983
always the last value, if R or B
14:45
BAndiT1983
blue requires top and left value, which should be interpolated while doing main processing, as they are later required for image border calculation
14:50
TofuLynx
what do you mean by top and left value?
14:50
BAndiT1983
check the RGGB pattern again
14:51
BAndiT1983
first value of B is at diagonal cross at 2,2 , so you need to calculate top and left values from available ones, ah and also right one, forgot about it
14:54
BAndiT1983
but the right value is the left of next blue calculation
14:54
BAndiT1983
sorry, can't do 2 things at once, trying to get the tax declaration done
14:54
TofuLynx
but that would do the same thing, you would be left with missing borders at bottom right and missing column and row next to those bottom and right borders
14:55
TofuLynx
no problem!
14:57
BAndiT1983
nope bottom and right would be automatically filled in, when iteration is at next value or at next row
14:57
TofuLynx
oops
14:57
BAndiT1983
you would calculate 2,2 then 1,2 and 2,1
14:57
TofuLynx
forgot to say that I mean that would happen when on BGGR pattern
14:58
BAndiT1983
it's the same, but for red
14:58
TofuLynx
so
14:58
BAndiT1983
there is no point to focus on colors, if they are not green values
14:58
BAndiT1983
just on cases and their starting points/ offsets
14:58
TofuLynx
how do I change the algorithm according to the pattern then?
14:59
BAndiT1983
add calculation for top and left for now
14:59
TofuLynx
hmm
14:59
TofuLynx
uuh
14:59
TofuLynx
i marked myself as away
14:59
TofuLynx
I dont know how xD
14:59
TofuLynx
Ok. sorry
14:59
TofuLynx
hmm
15:00
BAndiT1983
away? haven't seen any notification
15:00
TofuLynx
Oh well xD
15:00
TofuLynx
ok, so I see your point now
15:01
TofuLynx
by doing the blue channel as top and left and center
15:01
TofuLynx
I would have blue channels in the bottom and right borders, which would allow me to do bilinear interpolation
15:01
BAndiT1983
yep, or red channel, this are special cases, then this ones are at the end of the pattern, as end i mean like RGGB or GBBR
15:02
BAndiT1983
as the obviously suffer from missing sensels and have to be restored
15:02
TofuLynx
what happens in the mirror patterns, aka GBRG or GRBG?
15:02
TofuLynx
top and right in those cases?
15:02
BAndiT1983
G has more data available, so interpolation is rather simple,
15:03
BAndiT1983
you have to calculate it through weighting, there were already examples in the links we have exchanged some times ago
15:03
TofuLynx
yeah, for now I am doing simple weighting (dividing by four)
15:07
BAndiT1983
it's sufficient at the moment, adjustments will be done later
15:07
TofuLynx
Can you tell me, what pattern does axiom uses in fact?
15:07
TofuLynx
RGGB?
15:09
BAndiT1983
depends on the sensor
15:09
TofuLynx
CMV1200
15:09
BAndiT1983
don't know if it is always the same pattern
15:10
TofuLynx
Okk :)
15:10
supragya
depends on ordering too... if mirrored frames (bottom-up DIP), we get GBRG
15:11
aombk2
left the channel
15:11
supragya
last row => first row and so on
15:11
BAndiT1983
datasheet shows RGGB examples, but it could also be different, Bertl_oO or se6astian knows more
15:11
TofuLynx
Ok!
15:11
TofuLynx
I am just trying to understand why do the samples seem wrong
15:12
BAndiT1983
it's the conversion, look for raw2dng in the apertus repos
15:12
BAndiT1983
maybe we can fix it at some point
15:12
TofuLynx
hmm
15:13
TofuLynx
do you think that if I changed the tag of the pattern to RGGB or else, it would potentially be correct?
15:13
BAndiT1983
https://github.com/apertus-open-source-cinema/misc-tools-utilities/tree/master/raw2dng
15:13
BAndiT1983
can't say if it is just the flag or the order, so it needs some tests
15:14
TofuLynx
okk
15:14
BAndiT1983
it could be fead with RAW12 and then tested automatically, if the output is correct
15:15
BAndiT1983
but beware, a lot of code is in the source
15:19
TofuLynx
well It's not really a priority right now
15:19
TofuLynx
But it's something that has definitely to be investigated
15:21
ymc98_1
left the channel
15:21
ymc98_1
joined the channel
15:22
TofuLynx
BAndiT1983: Do you remember that sample with a girl in a green background?
15:27
BAndiT1983
yes
15:31
TofuLynx
in darktable, the color render correctly and such. but in OC it appears pink where it should be green
15:34
BAndiT1983
probably it has additional info, like linearization table
15:34
TofuLynx
I sent a image on lab chat
15:35
BAndiT1983
which pattern is recognized?
15:35
BAndiT1983
and by the way, tested with this image long time ago, had no problem
15:36
TofuLynx
It's recognized as GBRG
15:37
BAndiT1983
can you chec k it in darktable, maybe you can switch debayering off?
15:37
TofuLynx
I will try it
15:37
BAndiT1983
will also install newest darktable version+
15:38
BAndiT1983
nevermind, did already some days ago, getting old :D
15:38
TofuLynx
xD
15:39
BAndiT1983
image has pink tone, so green is diminshed and the skin is totally pink
15:40
TofuLynx
which is the opposite of what happens with OC
15:40
TofuLynx
also on OC the two green sensels have significantly different values, which introduces a grid artifact effect
15:40
BAndiT1983
hm, pity, you can switch to monochrome version, as passthrough, would prefer raw pattern in darktable, but ok
15:41
BAndiT1983
we should investigate it,l but if pattern is not correct, then this would explain different green values
15:41
TofuLynx
maybe it's an issue with TIFF loader (?)
15:41
BAndiT1983
TofuLynx: you have got exif info, was there something else related to CFA or channels?
15:42
TofuLynx
let me paste the tags
15:42
BAndiT1983
before we change code, get some DNG images from the web, which we can use for testing
15:42
TofuLynx
CFA Pattern : [Green,Blue][Red,Green]
15:42
TofuLynx
CFA Pattern 2 : 1 2 0 1
15:42
BAndiT1983
https://www.camerafv5.com/pages/raw-samples.php
15:44
BAndiT1983
ah, it's white balancing, if you set it to manual and drag a small rectangle on her shirt, then the image turns more natural
15:44
TofuLynx
hmm BAndiT1983
15:44
TofuLynx
i downloaded the beach picture on the page you sent
15:44
TofuLynx
and the image is very strange on OC
15:44
TofuLynx
I will send a picture
15:45
TofuLynx
Sent on lab
15:46
TofuLynx
do you see those vertical dark columns?
15:46
TofuLynx
well they are columns, of couse they are vertical xD
15:46
BAndiT1983
yep, probably preprocessor isn't handling this pattern yet
15:46
TofuLynx
Okk
15:47
TofuLynx
So, should I keep doing the bilinear Debayer with unit tests?
15:48
BAndiT1983
yes, it would allow to enclose problems
15:48
BAndiT1983
doing the same for daemon at the moment
15:48
TofuLynx
Ok :)
15:50
TofuLynx
Will be afk for a moment, I will eat something
15:50
BAndiT1983
have a nice meal
15:50
supragya_
joined the channel
15:51
supragya_
hey
15:51
BAndiT1983
hi again supragya
15:52
supragya_
who's there?
15:53
BAndiT1983
reminds me of the old french movie fantomas :D
15:56
supragya_
dude? I wanna talk to io. stream.h guy :P
15:57
supragya
hey supragya_
15:57
supragya
PM me please
15:57
supragya
not me BAndiT1983!
15:59
intrac
left the channel
16:03
intrac
joined the channel
16:03
intrac
left the channel
16:03
intrac
joined the channel
16:24
supragya
BAndiT1983: I guess g3gg0 will not be here today. We meet tomorrow for meet then
16:25
supragya
*(bad phrasing of sentence)
16:29
BAndiT1983
ok, then let's discuss it tomorrow
16:57
se6astian
changed nick to: se6astian|away
17:08
aombk
joined the channel
17:21
TofuLynx
left the channel
17:52
supragya_
left the channel
18:27
flesk_
left the channel
18:27
vup
left the channel
18:27
vup
joined the channel
18:27
anuejn
left the channel
18:43
anuejn
joined the channel
18:44
ZNC
left the channel
18:44
Kjetil
left the channel
18:44
Kjetil
joined the channel
18:44
ZNC
joined the channel
18:52
flesk_
joined the channel
19:15
BAndiT1983
changed nick to: BAndiT1983|away
19:17
BAndiT1983|away
changed nick to: BAndiT1983
19:32
se6astian|away
changed nick to: se6astian
19:48
BAndiT1983
changed nick to: BAndiT1983|away
20:02
TofuLynx
joined the channel
20:24
anuejn2
could anyone review the information added under https://wiki.apertus.org/index.php/AXIOM_Beta/Camera_Structure#Plugin_Modules_and_Shields ?
20:49
RexOrCine
changed nick to: RexOrCine|away
20:53
BAndiT1983|away
changed nick to: BAndiT1983
21:14
TofuLynx
BAndiT1983
21:14
TofuLynx
Are you there?
21:18
intrac_
joined the channel
21:20
intrac
left the channel
21:20
anuejn2
Bertl_oO: and whats the volteage range/ rules for VCC and VCC_IO on the shields?
21:22
anuejn2
can one just supply 3v3 to both, when one is not using lvds?
21:45
slikdigit
joined the channel
21:58
slikdigit_
joined the channel
21:59
slikdigit
left the channel
21:59
slikdigit_
changed nick to: slikdigit
21:59
se6astian
changed nick to: se6astian|away
22:12
TofuLynx
Good Night!
22:12
TofuLynx
left the channel
22:39
lexano
left the channel
22:42
BAndiT1983
changed nick to: BAndiT1983|away
22:42
BAndiT1983|away
changed nick to: BAndiT1983
23:12
slikdigit
left the channel
23:28
BAndiT1983
changed nick to: BAndiT1983|away
23:44
futarisIRCcloud
joined the channel