Current Server Time: 00:33 (Central Europe)

#apertus IRC Channel Logs

2015/11/25

Timezone: UTC


23:42
jucar
joined the channel
00:57
fsteinel
left the channel
00:57
fsteinel_
joined the channel
00:58
Bertl
changed nick to: Bertl_oO
03:45
aombk2
joined the channel
03:48
aombk
left the channel
05:27
LordVan
joined the channel
05:39
fsteinel_
changed nick to: fsteinel
05:54
Bertl_oO
off to bed now ... have a good one everyone!
05:54
Bertl_oO
changed nick to: Bertl_zZ
06:43
danieel
joined the channel
07:17
se6astian|away
changed nick to: se6astian
08:09
eliemichel
irieger_: do you have any documentation about this? To be sure I understand what you are talking about ;)
08:46
se6astian
irieger_ :Would love to see some raw as 1080p 12bit RGB packing for external raw recording. Easy task, isn't it ;-)
08:46
se6astian
if you refer to that
08:46
se6astian
I think it was meant as a joke
08:47
se6astian
its a mammoth project we will try to work on in the future
08:47
se6astian
for 4K 12bit raw though
08:47
se6astian
1080p 12bit RAW would be a bit easier... but still
08:48
se6astian
let me check how much we could in theory get over ethernet
08:49
se6astian
1920 x 1080 x 12bit = 2.966 Mbyte for one raw image
08:49
niemand
joined the channel
08:50
se6astian
so over gigabit ethernet we get a theoretical maximum throughput of 125Mbyte/s
08:50
se6astian
if there are no overhead related issues
08:50
se6astian
so that would mean we could get around 40 FPS with a raw stream
08:51
se6astian
this would require a streaming software running on Linux in the camera and some FPGA memory reading optimizations possibly
08:52
se6astian
maybe quite a big challenge for the first task to get warm with the camera
08:52
se6astian
but definitely something that would be of high value to the community
09:00
niemand
left the channel
09:16
aombk3
joined the channel
09:18
aombk2
left the channel
09:25
getzi
joined the channel
09:25
getzi
left the channel
09:26
getzi
joined the channel
09:54
eliemichel
se6astian: I suspected the sarcasm! Just that I still don't know what is done, what is still to be done and furthermore what are the references used in the project
09:54
eliemichel
so any doc is welcome ;)
09:57
eliemichel
For instance, are there like academic papers that you use at some point in the project? Or is every existing thing closed?
09:59
irieger_
eliemichel: Yeah, was a bit joking there. If it was possible to add a HDMI interface that is able to handle 1080p50/60 in full RGB 12 bit it may be possible to frame pack the raw 4K into 1080p using more than 1 frame. Not possible in the current hardware thought.
09:59
irieger_
I'm just nuts for the highest quality possible and "rawest" we can possibly get ;-)
10:00
irieger_
Just seen on twitter that the next team talk is nearly here. Don't tease us guys...
10:01
irieger_
changed nick to: irieger
10:03
irieger
Best thing of the apertus sensor btw.: It is 4:3 and could deliver nice 3672x3072 for anamorphic 4K shooting if we get this out of the cam somehow
10:05
eliemichel
may be a dumb question, but why cannot the images be slightly compressed, somehow (losslessly, I mean)?
10:06
eliemichel
Is it because there is some HDMI/whatever protocol to comply with?
10:06
eliemichel
Is it because it would take too much time?
10:06
eliemichel
Is it because we cannot be sure that it would work in any situation?
10:14
irieger
Don't know about the full capabilities of the FPGA but from the HDMI point: To be maximum compatible the best is to pack into a valid HDMI stream which means (afaik) have a image with RGB 8,10 or 12 bits or 422 (YCrCb) with alternating Y+Cr or Y+Cb per pixel with 10 bits (I think 8 may also be allowed)
10:15
irieger
if it is raw packe as a valid image it will not look like a picture so if a lossless compression is done before and then packed as a valid stream it shouldn't make a difference.
10:30
eliemichel
ok
10:31
Bertl_zZ
changed nick to: Bertl
10:31
Bertl
morning folks!
10:31
eliemichel
morning
10:34
Bertl
the problem is not to generate a valid HDMI stream containing raw data, the problem is more to make it a steam suited for a non raw recorder :)
10:38
eliemichel
yeah what is the protocol on this side?
10:39
Bertl
well, if the recorder does lossy compression for example, you need to produce an image which "looks" like a "normal" one, otherwise the compression will mangle your data heavily
10:41
Bertl
a simple, but inefficient example to do that is to double the frame rate (i.e. record 30FPS sensor data as 60FPS HDMI) an use alternating greens for example
10:41
eliemichel
so the decoder assumes it has some kind of image
10:41
eliemichel
in the stream
10:41
Bertl
i.e. first image is R/G1/B
10:41
Bertl
second frame is R/G2/B
10:42
Bertl
they will look like "normal" images but have the spatial information from the bayer matrix
10:42
Bertl
also inter frame changes will be minimal
10:42
eliemichel
bayer matrix?
10:43
Bertl
the sensor has a matrix of microfilters on top which are arranged like this
10:43
irieger
problem there: duplicating the values reduces the amount of data that can be packed.
10:43
Bertl
first row: RGRGRGRGRG ...
10:43
Bertl
second row: GBGBGBGB ...
10:43
eliemichel
ok
10:44
eliemichel
Bertl: this is what you call the bayer matrix, right?
10:44
Bertl
yes, that's what it is called :)
10:44
irieger
eliemichel: https://en.wikipedia.org/wiki/Bayer_filter
10:45
eliemichel
thx
10:45
Bertl
irieger: but on the other hand, the fact that always two consecutive frames are _very_ similar (only the green value changes so slightly) makes it easyy for the codec to compress
10:47
Bertl
as a bonus you get a better red and blue if you take it from the second frame
10:48
eliemichel
and how do people "usually" do?
10:49
irieger
Bertl: I wouldn't look at differences between the frames. No compression anyone who want's to get the best would be interframe. All the HDMI/SDI recorders out there are ProRes/DNxHD with intra-frame compression
10:50
irieger
Bertl: And why are the Reds and blues in the second frame better?
10:50
irieger
Btw. it's maybe best to get a good 422, 10bit out which most recorders understand and which is plenty of data to work with before doing the fancy stuff?
10:53
irieger
Meaning 422 already debayered of course.
10:55
Bertl
in case there is inter-frame compression, the reds and blues from the second frame will already have settled/improved
10:56
Bertl
sure, not a problem if there is only intra-frame compression
10:56
Bertl
eliemichel: do what?
11:06
irieger
Bertl: you know any recorders that do inter-frame?
11:08
Bertl
probably all recording devices which have very high compression, not saying that you will find them on the professional market though
11:09
irieger
Haven't seen any yet. Only recording devices doing H264 or so I know are all usb capture devices but no real recorders
13:07
aombk3
left the channel
13:19
aombk
joined the channel
13:44
eliemichel
Bertl: sorry, I got interupted
13:45
eliemichel
So I was asking: are they devices with raw 1080p 12bit RGB packing or something?
13:46
eliemichel
If so, somebody already though about it
13:49
Bertl
high end recorders are capable of recording raw data
13:49
Bertl
but basically any device which allows for uncompressed or lossless compression recording is suited for raw recording with the AXIOM
13:53
pozitrono
joined the channel
13:56
pozitron
joined the channel
13:59
pozitrono
left the channel
14:02
eliemichel
mmh ok
14:03
eliemichel
I mean, there seem to be some challenge here, which is to record that much data
14:03
Bertl
actually the raw data itself is not _that_ much if you think about it
14:03
eliemichel
have this very challenge already been addressed?
14:04
eliemichel
ok, then where's the tricky part?
14:04
Bertl
for example, if you take full HD at 36 or 48 bit (RGB)
14:04
eliemichel
the one that irieger mentioned with sarcasm =D
14:04
Bertl
you have 9331200 or 12441600 byte per frame
14:05
Bertl
now the native raw data from the cmv12k sensor is 4096x3072 in 8, 10 or 12 bit
14:05
Bertl
let's assume 12bit now, and you get 18874368
14:06
Bertl
in 8 bit, you already have 12582912, which is very close to the 48bit full HD
14:07
Bertl
the problem is partially to get the data out of the AXIOM Beta at resonable frame rates and to have a recorder which doesn't mess with the "raw" data
14:08
eliemichel
4096*3072*12 = 150994944
14:08
Bertl
you need to divide it by eight to get the bytes
14:08
eliemichel
my bad
14:09
eliemichel
ok it's getting much clearer in my mind =P
14:09
eliemichel
So my question was: Is it a somehow new problem, related to the axiom and its philosophy, or can other cameras do this?
14:10
Bertl
let's put it this way: there are many more options with a camera like the AXIOM, where you can control what data is generated and how
14:11
Bertl
proprietary cameras have certain modes (sometimes including something called "raw") to record data
14:11
Bertl
you can choose between them and that's it
14:12
Bertl
usually internal recorders are capable of recording more modes, as not all the modes are suitable for external transport
14:13
eliemichel
ok and then compression are mode-specific?
14:13
eliemichel
or maybe a given mode does not require all of the 4096x3072 values?
14:13
Bertl
yes, although the methods are quite similar regardless of the modes
14:16
Bertl
if you don't want to throw any data away, then you need to plan for the worst case, which means uncompressed
14:16
Bertl
in reality, you will be able to compress the data by 1/2 without losing any data (i.e. lossless)
14:19
eliemichel
yeah but the recorder need to uncompress, right?
14:19
eliemichel
or, to record losslessly
14:19
eliemichel
which sound like saving a lot of data…
14:19
Bertl
yes, I'm talking about the recording itself not the transport to the recorder
14:20
Bertl
I don't think there is a point in compressing the data _before_ you transport it to the recorder
14:20
Bertl
it might make sense if you transport data over e.g. gigabit ethernet though
14:20
eliemichel
well, if the transport capacity is limited, there is
14:20
eliemichel
I don't have the orders of magnitude in mind
14:21
Bertl
in theory, we have enough bandwidth on the output for all but the most extreme cases
14:21
eliemichel
se6astian was talking about GB ethernet
14:21
Bertl
we can do up to three Full HD (1080p60) HDMI ports on the plugin side
14:22
eliemichel
"the plugin side"?
14:22
Bertl
the AXIOM Beta has so called plugins for the output
14:22
LordVan
left the channel
14:22
eliemichel
it reminds me of something, I did not get it was for output
14:22
arpu_
joined the channel
14:23
Bertl
https://wiki.apertus.org/index.php/Beta_HDMI_Plugin_Module
14:23
arpu_
left the channel
14:23
Bertl
this is a single HDMI plugin module
14:23
Bertl
you can plug in two of them already, and we plan to do a 3xHDMI dual slot plugin module as well
14:24
eliemichel
thx
14:25
eliemichel
should explore more the wiki
14:41
getzi
left the channel
15:11
se6astian
changed nick to: se6astian|away
15:32
aombk
left the channel
15:33
aombk
joined the channel
15:59
irieger
Bertl eliemichel : Well compression could help. If about faktor 2 is possible it would be possible to use half the frame rate for raw. No need for 50/560p RGB in Deep color maybe?
15:59
irieger
meant 50/60p, not 50/560p of course
16:00
Bertl
the problem with lossless compression is always that you have to plan for the full bandwidth
16:02
irieger
Yep, haven't thought about this...
16:34
aombk2
joined the channel
16:36
aombk
left the channel
17:26
duvrai
left the channel
17:26
_florent_
left the channel
17:27
duvrai
joined the channel
17:27
_florent_
joined the channel
17:46
getzi
joined the channel
17:59
pozitron
left the channel
19:02
getzi
left the channel
19:04
getzi
joined the channel
19:27
getzi
left the channel
19:42
getzi
joined the channel
19:51
getzi
left the channel
20:01
getzi
joined the channel
20:12
getzi
left the channel
21:54
getzi
joined the channel
22:54
intracube_
joined the channel
22:55
intracube
left the channel
22:55
intracube_
changed nick to: intracube