Current Server Time: 15:05 (Central Europe)

#apertus IRC Channel Logs

2016/10/25

Timezone: UTC


00:07
BogdanXor
joined the channel
00:11
BogdanXor
left the channel
00:23
g3gg0
joined the channel
01:07
BogdanXor
joined the channel
01:12
BogdanXor
left the channel
01:22
Spirit532__
left the channel
02:08
BogdanXor
joined the channel
02:13
BogdanXor
left the channel
02:20
illwieckz
left the channel
02:21
illwieckz
joined the channel
02:51
arpu
left the channel
03:04
arpu
joined the channel
03:09
BogdanXor
joined the channel
03:14
BogdanXor
left the channel
03:34
jucar
left the channel
04:10
BogdanXor
joined the channel
04:14
BogdanXor
left the channel
05:11
BogdanXor
joined the channel
05:15
BogdanXor
left the channel
05:50
g3gg0
left the channel
06:11
BogdanXor
joined the channel
06:16
BogdanXor
left the channel
06:18
niemand
joined the channel
06:35
se6astian|away
changed nick to: se6astian
06:42
illwieckz
left the channel
06:43
illwieckz
joined the channel
07:06
niemand
left the channel
07:12
BogdanXor
joined the channel
07:17
BogdanXor
left the channel
07:30
Alex_Chooks
joined the channel
08:13
BogdanXor
joined the channel
08:18
BogdanXor
left the channel
08:23
ItsMeLenny
joined the channel
08:40
se6astian
changed nick to: se6astian|away
09:04
BogdanXor
joined the channel
09:33
niemand
joined the channel
10:10
Bertl_oO
alexML: I updated the firmware, in simulation it behaves exactly like the old one (when the GAIN LUTs are zero) but real world testing shows some weird clipping going on ... for now just disable the RCN clipping
10:11
Bertl_oO
fil_reg 11 0xF?41F000
10:11
Bertl_oO
where the 4 bits of ? are:
10:11
Bertl_oO
8 ... enable gain, 4 ... enable offset, 2 enable clip high, 1 enable clip low
10:12
Bertl_oO
so in theory ?=F should be the desired setup, but for now it seems that ?=D is required
10:15
Bertl_oO
building final stream now, will upload in a few minutes
10:26
Bertl_oO
http://vserver.13thfloor.at/Stuff/AXIOM/BETA/cmv_hdmi3_gain.bit
10:26
Bertl_oO
off to bed now ... have a good one everyone!
10:26
Bertl_oO
changed nick to: Bertl_zZ
10:33
niemand
left the channel
10:34
kaiserlich
joined the channel
12:14
se6astian|away
changed nick to: se6astian
12:21
Spirit532__
joined the channel
12:29
ItsMeLenny
left the channel
13:05
niemand
joined the channel
13:06
niemand
alexML, we decided not to use the axiom today
13:56
niemand
left the channel
14:07
kaiserlich
left the channel
14:19
niemand
joined the channel
15:30
toxitobi
joined the channel
15:31
toxitobi
good evening
15:35
se6astian
hello!
15:43
jucar
joined the channel
15:54
niemand
left the channel
16:11
niemand
joined the channel
16:19
kaiserlich
joined the channel
16:25
jucar
left the channel
16:30
niemand
left the channel
16:43
kaiserlich
left the channel
16:53
Spirit532
joined the channel
16:57
Spirit532__
left the channel
18:28
alexML
Bertl_zZ: the latest firmware gives correct image with any value for the "?"
18:29
alexML
however, there is a strange overlay at startup (also present with the old firmware) that disappears after a few attempts to change fil_reg 11
18:30
alexML
there is still a "ghost" of that overlay when running mimg commands
18:32
alexML
"./mimg -a -o /opt/IMG/newsmoke.rgb16" runs an animation on the left side of the screen
18:32
alexML
and, in the end, the overlay disappears
18:39
alexML
correction: the behavior of that "?" does not depend only on its value, but also on its history of values
18:40
alexML
e.g. setting it from 0 to F (trying each value) after reboot gives the following:
18:42
alexML
default: white screen; 0: blue screen with a small white line in the top-left corner; 1: no change; 2: the white line disappears;
18:43
alexML
3...7: no change; 8: normal image; any value after that: no change (still normal image)
18:46
DGMurdockIII
joined the channel
18:46
alexML
starting with D after reboot gives a white rectangle (thin line, not filled) over about 1/3 of the screen (left side)
18:47
alexML
changing to F => the rectangle disappears
18:47
alexML
back to D => the rectangle does not reappear
18:48
alexML
any value after that (0-F) => no change (normal image)
18:48
alexML
so, it's a state machine, not very easy to reverse engineer :P
19:00
BogdanXor
left the channel
19:04
kaiserlich
joined the channel
19:24
Bertl_zZ
changed nick to: Bertl
19:24
Bertl
back now ...
19:27
alexML
also, forgot to mention: I wasn't able to change the gains
19:28
alexML
(script linked yesterday)
19:28
Bertl
definitely doesn't happen here
19:28
alexML
do you remember which overlay is supposed to show the axiom beta logo?
19:29
Bertl
./mimg -a -o /opt/IMG/overlay_04.rgb
19:30
alexML
this adds the white rectangle I've described earlier
19:31
alexML
05 makes it a bit thicker
19:31
alexML
03 makes it disappear
19:31
Bertl
yep, that's expected ... it is how the overlay looks
19:31
Bertl
but the data is not in the sensor data
19:32
alexML
ah, you changed the meaning of the overlay to encode pixel offsets?
19:32
Bertl
not yet
19:32
Bertl
regarding overlay, everything should be as before with the new bitstream
19:33
alexML
yes, the problem is also present with the old bitstream (cmv_hdmi3_60.bit)
19:33
Bertl
show me what you get in a snap or see on HDMI, then I can tell
19:33
alexML
hold a second
19:39
alexML
https://dl.dropboxusercontent.com/u/4124919/bleeding-edge/apertus/overlay-issue.jpg
19:39
alexML
this is with overlay_04
19:39
Bertl
very interesting
19:40
Bertl
can you upload the .rgb for me?
19:40
alexML
overlay_05 just makes the border thicker
19:40
alexML
I didn't change it
19:40
Bertl
md5sum?
19:40
alexML
c47ddc7b7607ad0623c9bdf15fbfd2d2 /opt/IMG/overlay_04.rgb
19:40
Bertl
and the mimg tool?
19:40
alexML
8547e3c3b7abd98094eddae7baa7c07e /opt/IMG/overlay_05.rgb
19:41
alexML
653c33e76b5dabbeb40a4c400c61facd mimg
19:41
Bertl
that is definitely different from any I have
19:42
Bertl
let me quickly upload my version just to verify
19:42
alexML
this is the image you sent me a while ago to unbrick the beta
19:43
Bertl
it might have changed in the meantime, so I do not consider that a problem right away, we should just check it
19:43
Bertl
http://vserver.13thfloor.at/Stuff/AXIOM/BETA/mimg
19:43
Bertl
443840323893eb56171fb8b160ee9615 mimg
19:44
alexML
which: no mimg in (/usr/local/sbin:/usr/local/bin:/usr/bin:/usr/bin/site_perl:/usr/bin/vendor_perl:/usr/bin/core_perl)
19:44
alexML
what the duck?
19:45
Bertl
logout and login again, bash caches
19:45
Bertl
an use ./mimg
19:45
Bertl
*and
19:45
Bertl
(after changing the permission for it to be executeable)
19:46
alexML
I used that before too (copied from kick_manual)
19:46
alexML
works now
19:46
alexML
got the axiom beta logo
19:46
Bertl
okay, so I conclude that your mimg has some 'other' features
19:47
Bertl
maybe check with /opt/BERTL/mimg.c
19:47
alexML
368bea2b37cdb85c290c5883e8d9b006 /opt/BERTL/mimg.c
19:48
alexML
** Version 1.8
19:48
Bertl
(yep, md5 matches)
19:48
Bertl
ah, but that one indeed shows your picture
19:49
alexML
hm, maybe I screwed it up when trying to recompile cmv_snap3
19:49
DGMurdockIII
left the channel
19:50
Bertl
let me figure out why the mimg.c is actually broken
19:51
alexML
recompiled from /opt/BERTL (make clean; make) and ran that one
19:51
Bertl
the 'working' one is v1.6
19:51
alexML
result: the overlay is no longer transparent
19:51
alexML
and calling it multiple times mixes all the overlays
19:51
Bertl
yeah, avoid it for now, I'll figure out what got broken
19:51
alexML
e.g. overlapping "AXIOM \Beta" and "AXIOM Overlay"
19:52
alexML
but at least the overlay is full-screen
19:53
alexML
btw, looks like I've broken cmv_snap3.static when I've added the semaphore calls
19:53
alexML
for some reason, it can't link with -static -lpthread
19:55
alexML
found the answer: it must be -static -pthread (no idea why though): http://stackoverflow.com/questions/17928260/gcc-static-linking-error-for-lpthread
19:56
BogdanXor
joined the channel
20:01
BogdanXor
left the channel
20:03
alexML
anyway, do you mind reminding me of the address of the LUT entries? (after matrix)
20:03
Bertl
the gamma LUTs?
20:03
alexML
yes
20:04
Bertl
0x80300000, 0x80304000, 0x80308000, and 0x8030C000
20:04
alexML
one for each output channel, after matrix?
20:04
Bertl
correct
20:05
alexML
thanks
20:05
Bertl
np
20:06
alexML
ah, found lut_conf3
20:07
Bertl
yep, it is used in gamma_conf.sh
20:07
alexML
right
20:07
alexML
btw, since you are here
20:08
alexML
talked with se6astian a bit about recording the black columns on the HDMI (for the experimental raw mode)
20:08
alexML
currently they are not used in this mode
20:08
Bertl
okay?
20:08
alexML
and here I have some figures showing what improvement they can give
20:09
alexML
https://wiki.apertus.org/index.php/Pattern_Noise#Tests_on_a_dark_frame
20:09
alexML
short summary:
20:09
alexML
after dark frame, row noise stdev is 1.33
20:10
alexML
after subtracting black columns with a scaling factor, it goes down to 0.82
20:10
alexML
if I do some more complex processing (including looking at green channel imbalance), I can bring it down to 0.31
20:11
alexML
so, the info from those columns can be used to make the row noise 4 times smaller
20:11
Bertl
sounds great
20:11
alexML
that means, I think we should record this info somehow
20:11
alexML
but also keep it simple on the FPGA side
20:12
alexML
what about averaging the 16 black columns (for each row), and outputting the result on one border column on the HDMI image?
20:12
Bertl
tricky
20:13
Bertl
the problem is, the black rows are not fetched for the HDMI output
20:13
Bertl
*columns
20:13
Bertl
so the data is not available in the output pipeline
20:13
alexML
hm
20:14
alexML
you effectively crop the image, right?
20:14
Bertl
yes, the output pipeline is an address generator, which selects a 4x1080p data frame (or 1080p 4 sensel area)
20:16
Bertl
it might be possible though, to average (or sum) the data on the arm cores and save it separately
20:18
Bertl
we are talking about 1.5 megabit of 'raw' data here per frame
20:18
Bertl
so roughly over one megabyte per second
20:19
Bertl
if summed/averaged, it will be reduced by roughly a factor of 16
20:19
alexML
can you point me to a way to sync the arm code with the image capture? (interrupt or whatever)
20:20
Bertl
the snap code waits for a buffer change and extracts the currently displayed frame address
20:20
jucar
joined the channel
20:20
alexML
alright
20:21
alexML
and some address where I can write data to appear on the HDMI output?
20:21
Bertl
so the easiest way is to map all frame buffers and wait for a register to change
20:21
Bertl
once the buffer is on display, it is guaranteed not to change
20:21
Bertl
and it is also guaranteed to be complete at the moment
20:21
alexML
it can be the overlay as well
20:22
Bertl
yes, you can always write to the overlay
20:22
alexML
and having the data shifted by one frame shouldn't be a big deal (since it's a video stream, after all)
20:22
Bertl
if you are fast, you can even write the data on the same buffer
20:22
Bertl
let's say your code takes 500 lines to do the calculations
20:23
Bertl
and you have an uncertainty of 250 lines for the start
20:23
Bertl
that would leave you about 2000 lines to construct and write an overlay for the bottom line
20:35
Alex_Chooks
left the channel
20:42
se6astian
off to bed
20:42
se6astian
gnight!
20:42
se6astian
changed nick to: se6astian|away
20:57
BogdanXor
joined the channel
21:01
BogdanXor
left the channel
21:47
alexML
Bertl: I have some trouble with clipping, for example: ./mat4_conf 0 0 0 0 0 0 0 2.8 0 0.7500 0.7500 0 1 0 0 0
21:48
alexML
ideally, the stuff that overflows should clip to white
21:48
alexML
instead, I get bright red and magenta patches (I can take a screenshot if you need)
21:49
Bertl
well, clipping is per channel, i.e. it clips the channel to the maximum value
21:49
Bertl
doesn't affect the other channels though
21:50
Bertl
but please show we what you see and what you expect
21:56
alexML
https://dl.dropboxusercontent.com/u/4124919/bleeding-edge/apertus/IMG_7311.jpg
21:56
alexML
https://dl.dropboxusercontent.com/u/4124919/bleeding-edge/apertus/IMG_7312.jpg
21:57
Bertl
that looks like the clipping is disabled
21:57
alexML
what happens is that blue wraps back to 0
21:57
alexML
ok, how to enable it?
21:57
alexML
fil_reg 11 0xF?41F000 has no effect for any value of ?
21:57
alexML
(see the state machine described above)
21:58
Bertl
fil_reg 11 0xFC?1F000
21:58
BogdanXor
joined the channel
21:58
Bertl
did I paste the wrong place? ... that would explain the overlay issue
21:59
Bertl
bits 20-21 are the clipping, bits 22-23 the enable bits
21:59
alexML
12:11 < Bertl_oO> fil_reg 11 0xF?41F000
21:59
Bertl
bits 24-31 are the write mask for bursting the image into memory
22:00
Bertl
they are usually 0xFC, i.e. they leave out the lower 16 bits
22:01
Bertl
(the overlay)
22:01
Bertl
but as I said, there is something wrong with the upper bound clipping on the new firmware
22:02
alexML
ok, let me roll back
22:02
Bertl
I can upload the source if you like play with that
22:02
Bertl
simulation says everything is fine and gives the expected results, but, as so ofen, this doesn't match reality
22:03
Bertl
*often
22:03
BogdanXor
left the channel
22:05
alexML
back to old firmware, but still the clipping flag doesn't seem to get enabled
22:05
alexML
default value for fil_reg 11 is 0xFC31F000
22:05
Bertl
note that this is for the RCN clipping
22:05
Bertl
there is another clipping for the matrix
22:05
alexML
ah
22:06
Bertl
https://wiki.apertus.org/index.php/AXIOM_Beta_Software#Clipping
22:09
alexML
perfect, works as expected now, thanks
22:09
Bertl
you're welcome!
22:12
kaiserlich
left the channel
22:19
toxitobi
left the channel
22:20
alexML
Bertl: for mat4_conf.sh, can the offset be negative?
22:21
alexML
./mat4_conf.sh 0 0 0 0 0 0 0 14.0000 0 3.7500 3.7500 0 5.0000 0 0 0 -0.1 -0.1 -0.1 -0.1
22:21
alexML
I get the same output as with ./mat4_conf.sh 0 0 0 0 0 0 0 14.0000 0 3.7500 3.7500 0 5.0000 0 0 0 0.1 0.1 0.1 0.1
22:21
alexML
but with 4 warnings: dc: stack empty
22:21
alexML
(repeated 4 times)
22:24
alexML
found out that it wraps at 16
22:24
alexML
so 16 is the same as 0, and 15 is -1
22:24
alexML
now I'm not sure what the units are
22:27
alexML
0.5 maps - very roughly - to about 200 12-bit DN
22:36
alexML
wait, offset is applied after multiplication?
22:37
Bertl
I suggest to use the mat4_conf.py instead
22:37
Bertl
it handles negative values as well as 3x3 matrices
22:37
alexML
yep, looks like
22:43
alexML
so, at gain 1, 15.89 = -0.11 maps to about -500
22:43
alexML
500/4095 = 0.12, so 1.0 = 4095?
22:45
Bertl
matrix or LUT?
22:45
alexML
I have trouble finding mat4_conf.py (neither google "mat4_conf.py" nor "locate mat4_conf.py" can find it)
22:45
alexML
matrix
22:47
Bertl
http://vserver.13thfloor.at/Stuff/AXIOM/BETA/mat4_conf.py
22:48
alexML
gracias
22:48
Bertl
np
22:49
alexML
syntax is different?
22:50
Bertl
yes and no, it should be compatible for 16/20 values
22:51
Bertl
but the order is reversed
22:51
Bertl
i.e. previously the order was B G2 G1 R now it is R G1 G2 B (as one would expect)
22:51
Bertl
similar to R G B for the 3x3 case
22:52
alexML
got it
22:52
alexML
so I'm now able to apply the matrix with some arbitrary black level as origin
23:00
BogdanXor
joined the channel
23:04
BogdanXor
left the channel
23:12
alexML
ok, you may review the API: https://dl.dropboxusercontent.com/u/4124919/bleeding-edge/apertus/set_wb.py
23:13
alexML
still WIP, but setting WB as RGB multipliers appears to work
23:14
Bertl
well, you can use mat4_conf.py rgain ggain bgain rblack gblack bblack
23:14
Bertl
it will do the mapping to the 4x4 matrix for you
23:15
alexML
yes, but I want to set a full matrix in the end, not just gains
23:16
Bertl
I also think you are probably better off importing the mat4_conf.py parts like niemand did with the HSL stuff
23:17
Bertl
http://sebix.at/tmp/rgbhsv.py