Current Server Time: 13:37 (Central Europe)

#apertus IRC Channel Logs

2015/12/05

Timezone: UTC


23:14
arpu
left the channel
23:55
pozitron
left the channel
00:42
Bertl_oO
off to bed now ... have a good one everyone!
00:42
Bertl_oO
changed nick to: Bertl_zZ
02:31
intracube
irieger: interesting link, thanks
05:02
pozitrono
joined the channel
06:40
pozitrono
left the channel
07:33
pozitron
joined the channel
08:23
Bertl_zZ
changed nick to: Bertl
08:23
Bertl
morning folks!
08:43
intracube
morning Bertl :)
08:44
intracube
was the recent test footage run through motion interpolation software
08:44
Bertl
not that I'd know of, why?
08:45
intracube
can see some glitches that don't look like regular artefacts from debayering or regular processing
08:45
intracube
https://youtu.be/izDzFov3nDM?t=208
08:45
Bertl
well, I'm pretty sure youtube does their own recompression thing
08:46
intracube
it doesn't look like youtube compression
08:46
intracube
the side of the football has a weird patch on it
08:46
intracube
also a few frames later there's some even more weird stuff
08:46
intracube
will post a screencap in a mo
08:47
Bertl
you probably want to check with the raw data and/or se6astian|away when he wakes up
08:47
intracube
yup ok :)
08:52
intracube
couple of sample frames for se6astian : https://goo.gl/photos/QYggaXmfiQzKZpUK9 https://goo.gl/photos/zrMwj77c53dJVAft9
08:55
Bertl
doesn't even remotely look like data from the raw capture, so I'm pretty sure that was motion encoded (wherever that happened :)
08:58
comradekingu
Emulsion film?
09:03
Bertl
you can't go around and tell everybody about our secret :)
09:34
intracube
Bertl: yep, also some strong colour correction (for effect rather than correcting for sensor issues)
09:34
intracube
even though the video said footage was raw with no correction or compensations
09:34
intracube
which could be misleading
09:36
Bertl
as I said, no idea how it was processed, in any case I would check out the raw data, which should be available somewhere (se6astian will know)
09:37
intracube
yep
09:40
Bertl
from my tests, everything looks very similar to the alpha, the dynamic noise is definitely lower (probably due to the better design of the power supply)
09:40
intracube
I've had a quick look at the DNG files and they look neutral except for the pink highlights
09:40
Bertl
we have those strange artefacts on the right end of the sensor, which are unexplained yet (I suspect sensor register issues)
09:42
Bertl
colored highlights might be due to near IR or just the sensor profile (I presume the raw data is within the 12bit range)
09:43
Bertl
so a tint there is not unusual
09:43
intracube
have cmosis given an explanation for the artefacts on the right?
09:44
Bertl
not that I know of, but I'm unsure if it was properly reported yet
09:44
intracube
ah
09:44
Bertl
fact is, we have been seeing this on at least two different sensors (same model)
09:45
Bertl
we will shortly test with the engineering sample we used on the alpha, to figure out if it is caused by the new sensors (v2) only
09:46
Bertl
but to be honest, it could be caused by a lot of subtle changes done on the beta, so it definitely needs some more investigation
09:47
intracube
troy_s mentioned that the pink highlights was a known issue and this data would probably have to be clipped/ignored?
09:47
intracube
I noticed something similar on my Nikon DSLR when I enabled highlight recovery
09:47
intracube
(on the RAW software converter)
09:48
intracube
the image clearly had more highlight data, but it was skewed to pink/purple
09:48
intracube
istr that normally that data is ignored when outputting JPEGs or when converting RAW with highlight recovery disabled
09:49
Bertl
yes, of course, but this is to be expected
09:50
Bertl
"proper" raw data should not reach the lowest and highest values on each channel (for 12bit that is 0 and 4095)
09:50
intracube
it can be useful data in an emergency where footage would otherwise have to be junked
09:51
Bertl
otherwise the offset or gain is set wrong and clipping is to be expected
09:51
intracube
https://goo.gl/photos/JmwLbQMw2QqXgSmS9 (left: sample DNG, right: processed in Blender)
09:52
Bertl
yes, that's a good example, on the left, you see that neither black nor white is reached
09:55
pozitron
left the channel
10:04
pozitrono
joined the channel
10:34
alexML
Bertl: the .raw12 file for that sample has lots of clipped blacks; in the DNG converter, I added an option to artifially raise the blacks in order to reduce color casts, but the sensor offset was way too low
10:35
alexML
(I'm talking about confettiTV-32.raw12, which appears to be from the same movie as intracube's link)
10:36
alexML
the "dark magenta" background from the left image is from artificially raising the black level
10:45
intracube
alexML: ah yep. I noticed the crushed blacks. I was expecting to see FPN and sensor noise in the darkest parts of the frame
10:45
intracube
but it was completely absent
10:46
alexML
yes, it's clipped
10:46
alexML
I wanted to see it as well, as I'm experimenting with FPN correction
10:54
intracube
I noticed similar black clipping on some of the footage from the alpha camera
10:59
Bertl
alexML: yeah, I think the reason for that is that with the live view, folks tend to make the blacks really black, but with the wrong adjustments
11:00
Bertl
i.e. either with the sensor registers or with the linearization lut, instead of the matrix or gamma lut (where it should happen)
11:41
irieger
intracube Bertl: You talk about the artifacts like in https://youtu.be/izDzFov3nDM?t=236 when meaning the artifacts on the right?
11:41
irieger
I
11:41
irieger
I'd take a "wild" guess and say some sensor params(read registers) could be tuned to match the used readout mode etc ;-)
11:42
irieger
alexML: what experiments do you do with fpn reduction?
12:07
Bertl
maybe, but we should have the "defaults" we used on the alpha
12:08
Bertl
nevertheless, more testing needs to be done
12:09
irieger
Yep, testing is always needed with this stuff. But to assure people I'd say it's not the sensors, it's the tuning of the sensor params and the software I'd say
12:11
alexML
irieger: I'm trying to find a way to estimate the FPN from any real-world image, since the optical black areas are way too small on this sensor
12:11
alexML
testing data (sample DNGs) welcome :P
12:12
irieger
Real world data? How about images of black? ;-)
12:12
irieger
I know, they don't show everything...
12:12
alexML
:P dark frames are very easy to fix, I already analyzed them in this thread: http://www.magiclantern.fm/forum/index.php?topic=11787
12:16
irieger
alexML: interesting. Have to look through this. So you (off course I knew who you are) took the stuff over to your forum. Bad guy ;-)
12:17
irieger
Thanks for ML btw. You are the reason I switched to Canon with their inferior sensor instead of Nikon when I jumped to Sony ship when it was clear they wouldn't make any more DSLRs ...
12:52
arpu
joined the channel
14:16
intracube
highlight recovery + neutral grade test: https://www.youtube.com/watch?v=GTQzGedoxrc
14:58
se6astian|away
changed nick to: se6astian
15:14
troy_s
The transform should maintain some possible negative values for sensor noise.
15:16
troy_s
God I can't read that thread.
15:23
troy_s
intracube: Remember too (not that I know anything about the Beta, as I haven't worked on it yet) that the linear portion of the sensor doesn't carry through the full range of data (hence why when doing HDR merges software keeps a solid anchor in the linear portion). Any colour transform has to take this into account and nonlinearly adjust the data (hence 3D
15:23
troy_s
LUT).
15:24
troy_s
intracube: Most transforms vendors use are strictly a matrix, which is 'good enough'. A proper colorimetric transform for post production would ideally hold the proper XYZ values across the full range of data, and as a result, be a nonlinear transformation to avoid that colour casting you are seeing near the highlights. Hello rat piss yellow.
15:29
se6astian
changed nick to: se6astian|away
17:30
mithro
left the channel
17:31
mithro
joined the channel
21:03
pozitrono
left the channel
22:16
rbckman
joined the channel
22:51
rbckman
left the channel