23:00 | Bertl | the alpha prototype was designed to allow full HD @ 25 FPS and maybe take 4k stills
| |
23:01 | danhanes | Makes sense - seems like pretty much the limit of the Zinq.
| |
23:01 | Bertl | we already have the 4k stills working perfectly as well as the HDMI output (although not of live images)
| |
23:01 | Bertl | the Zynq can do a lot more, but we are limited by the Zedboard peripherials
| |
23:02 | Bertl | i.e. the Zynq can easily handle 75FPS at 4k in the current setup (capture)
| |
23:02 | danhanes | Right - mixed terminology here :)
| |
23:02 | Bertl | and if done correctly, the 150FPS shouldn't be a problem either
| |
23:02 | Bertl | (at 12bit, full 4k that is)
| |
23:03 | Bertl | the problem is more the output, as the 7020 doesn't feature any multi gigabit tranceivers
| |
23:03 | danhanes | I believe you mentioned moving to the 7030 at one point
| |
23:04 | danhanes | Still capable of webpack development
| |
23:04 | Bertl | yes, it will be either 7015 or 7030 for the next step
| |
23:05 | danhanes | Once I feel competent to contribute some code - what area would is be most helpful for me to focus on ?
| |
23:06 | Bertl | depends on where we are at that point, I guess?
| |
23:07 | danhanes | right
| |
23:16 | dmj_nova | joined the channel | |
00:11 | danhanes | left the channel | |
00:32 | gcolburn | left the channel | |
00:38 | danhanes | joined the channel | |
01:50 | danhanes | left the channel | |
02:09 | Rambutan2000 | left the channel | |
02:17 | Rambutan2000 | joined the channel | |
02:17 | Rambutan2000 | Morning all.
| |
02:18 | dmj_nova | left the channel | |
02:37 | Bertl | morning Rambutan2000!
| |
02:39 | dmj_nova | joined the channel | |
03:13 | Rambutan2000 | I've been thinking about the design of the case and what features would be good bad. Since this project is open source do you guys think being able to 3d print and/or CNC parts of the case is important?
| |
03:14 | Rambutan2000 | Hi Bertl!
| |
03:20 | Bertl | well, it might not be a primary goal atm, but I guess it would be nice to be able to fabricate certain parts yourself
| |
03:23 | Rambutan2000 | Also has anyone been thinking into bus design for the expandable platform? I've seen open source IP cores for PCIe, but I've also been reading how USB 3.1 can reach 10gb/s
| |
03:23 | Rambutan2000 | http://www.youtube.com/watch?v=isQ7cvuyoTw&feature=c4-overview-vl&list=PL1AFF5D9EB822C193
| |
03:24 | Bertl | USB is nice for interoperation, but as protocol, it is needlessly complicated
| |
03:25 | Bertl | e.g. to implement a serial console in USB 1.1 (not even 2.0) you need a lot of hardware and software, but you can achieve the same with almost no resources using rs232
| |
03:25 | Rambutan2000 | yeah I tried reading through the 3.1 spec
| |
03:26 | Bertl | with higher data rates and newer versions the number of features increases, the hardware gets more expensive and protocols get more complicated
| |
03:26 | Bertl | and we definitely do not want a proprietary stack in our design
| |
03:28 | Rambutan2000 | nope
| |
03:54 | Rambutan2000 | Also what's the current thoughts on storage? SSD RAID?
| |
04:00 | Bertl | first step will be SDI with external recorder, i.e. no storage at all
| |
04:00 | Bertl | second step will be SSD raid on small SATA 6G SSDs
| |
05:20 | Bertl | off to bed now ... have a good one everyone!
| |
05:39 | jucar | joined the channel | |
06:33 | jucar | left the channel | |
07:21 | jucar | joined the channel | |
09:27 | jucar | left the channel | |
09:51 | jucar | joined the channel | |
10:19 | jucar | left the channel | |
10:56 | danhanes | joined the channel | |
11:41 | danhanes | left the channel | |
12:58 | philippej | joined the channel | |
13:05 | Bertl | morning everyone!
| |
13:06 | philippej | Morning Bertl !
| |
13:06 | philippej | How goes?
| |
13:07 | Bertl | slow but steady ... and on your end?
| |
13:07 | philippej | Quiet and sleepy :-)
| |
13:09 | philippej | any news for image calibration ?
| |
13:13 | Bertl | well, actually we did some calibration tests yesterday and I have written a script to create a perfect IT8 chart based on the sensor data
| |
13:13 | Bertl | it just needs somebody to run it through the various tools
| |
13:15 | Bertl | http://vserver.13thfloor.at/Stuff/AXIOM/it8-tungsten-01-12ms-perfect.png
| |
13:16 | Bertl | I'm curious how that one will score with those tools
| |
13:22 | philippej | impressive, how have you made it?
| |
13:23 | Bertl | bash, gawk, imagemagick
| |
13:23 | philippej | :-)
| |
13:24 | S3bastian | joined the channel | |
13:24 | S3bastian | hello!
| |
13:25 | philippej | Hello !
| |
13:38 | philippej | I have agryll runing, but I'm not 100% sure I know what I'm doing :-)
| |
13:39 | philippej | using your latest png, I get average error 5.664896 at the end of the chain
| |
13:39 | philippej | problem is that I used some random reference file and don't know if all the command line arguments are right
| |
13:40 | Bertl | which is quite interesting, as the chart itself is really perfect :)
| |
13:40 | philippej | don't you think it's the error of what the tool can achieve given those colors?
| |
13:41 | Bertl | I have no clue what the average error means (and I've posed the question several times without getting any answer)
| |
13:41 | guesst | what color space is the image ? :) (one should put that in filename for clarity)
| |
13:42 | philippej | here some info : http://comments.gmane.org/gmane.comp.graphics.argyllcms/5046
| |
13:42 | guesst | to me, the grays are not gray but yellow...
| |
13:42 | Bertl | which is kind of expected with tungsten sources
| |
13:43 | Bertl | note that this is not an idealistic chart, it is a perfect version of the actual sensor taking a picture of the chart
| |
13:53 | philippej | at least I get an icm file
| |
13:54 | philippej | I think the error % is based on what the tool managed to produce with the color it received
| |
13:55 | Bertl | we'll see what the calibration experts make from that chart
| |
13:56 | philippej | at least when I apply it to the png file in photoshop, I get an image that looks much more like a good chart
| |
13:57 | philippej | here : http://public.philippejadin.be/it8-tungsten-01-12ms-perfect-with%20profile%20applied.png
| |
13:58 | Bertl | yup, doesn't look completely wrong
| |
13:59 | philippej | but note I used a random reference file for IT8. I'd need the right reference file, for the precise batch used when printing the chart, as I understand it. Anyway, I guess experts will know better and faster :-)
| |
13:59 | Bertl | IMHO the chart was a little overexposed which shows in the bright areas
| |
14:00 | philippej | that's what those percentage of error mean
| |
14:00 | S3bastian | philippej: try this chart reference file: www.colorreference.de/targets/R100604.zip
| |
14:01 | philippej | S3bastian
| |
14:01 | philippej | doing right now
| |
14:02 | S3bastian | great
| |
14:07 | Bertl | S3bastian: btw, the chart image shows quite a number of speckles probably caused by dirt on the sensor
| |
14:08 | Bertl | (or maybe dirt on the chart? :)
| |
14:08 | S3bastian | good to know, I will clean the sensor and try to keep the new chart as clean as possible when it arrives
| |
14:08 | Bertl | 10A, 12C, 16C, 19C foe example
| |
14:09 | Bertl | *for
| |
14:09 | S3bastian | in your chart script do you also compensate for FPN already?
| |
14:09 | Bertl | 11H, 12F, 4F, 2L, 13H and so on and so forth
| |
14:10 | Bertl | no compensation so far, just statistics
| |
14:11 | philippej | weird result, probably caused by weird manipulations :-) http://public.philippejadin.be/it8-tungsten-01-12ms-perfect-with-profile-applied-bis-ok.jpg
| |
14:13 | Bertl | what I wonder is if the tools account for differences in brightness over the chart or if they simply assume it is evenly lit
| |
14:13 | S3bastian | do you think the grey "haze" from FPN could be an issue for the calibrations or will it not alter saturation levels?
| |
14:14 | Bertl | because the image I used has quite some differences between left and right (and also top to bottom but not so much there)
| |
14:14 | Bertl | i.e. it would be really beneficial to take a picture of the chart which fulfills the following requirements:
| |
14:14 | Bertl | - no reflections
| |
14:14 | Bertl | - even lighting (two light sources, one on each side)
| |
14:15 | Bertl | - no over or underexposure
| |
14:15 | Bertl | the actual angle doesn't matter much for this approach
| |
14:16 | Bertl | and of course, a number of images with the same setting would be nice to have to eliminate the dynamic noise pattern
| |
14:20 | gcolburn | joined the channel | |
14:23 | Bertl | off for now ... bbl
| |
14:27 | philippej | Hi gcolburn
| |
14:27 | philippej | if you read the backlog, you'll find a very interesting file to play with, Herbert's corrected png :-)
| |
14:32 | gcolburn | hello
| |
14:33 | gcolburn | yeah I just ran it through Argyll
| |
14:33 | gcolburn | 14.5% error unfortunately
| |
14:33 | philippej | is it bad?
| |
14:33 | S3bastian | I would like to understand what this error means exactly
| |
14:34 | philippej | latest try, profile applied, I cheated with the level feature to have stronger blacks : http://public.philippejadin.be/it8-tungsten-01-12ms-perfect-with-profile-applied-ter.jpg
| |
14:34 | gcolburn | I'll get you the best description I have found of it
| |
14:34 | dmj_nova | left the channel | |
14:34 | gcolburn | I had one chart give me 12.5% yesterday
| |
14:35 | gcolburn | I embedded the matrix in the DNG, and compared the result with the previous matrix
| |
14:35 | gcolburn | the images were visually indistinguishable. and those were comparing new chart captures with older ones
| |
14:35 | S3bastian | is the error the "distance" from what the colors should look like to what they do look like after applying a LUT for RGB each?
| |
14:35 | gcolburn | "If the -v flag is used (verbose), then at the end of creating a profile, the maximum and average fit error of the input points to the resulting profile will be reported. This is a good guide as to whether things have gone smoothly in creating a profile. Depending on the type of device, and the consistency of the readings, average errors of 5 or less, and maximum errors of 15 or less would normally be expected. If errors are grossly higher than this, th
| |
14:35 | gcolburn | this is an indication that something is seriously wrong with the device measurement, or profile creation."
| |
14:36 | gcolburn | from the documentation
| |
14:36 | gcolburn | the canon shot you sent me before was was 3.x%
| |
14:36 | philippej | I got this : Profile check complete, peak err = 17.696363, avg err = 3.949152, is it so bad you think ?
| |
14:37 | gcolburn | that's using Argyll and a new chart?
| |
14:37 | gcolburn | from the CMV12000?
| |
14:37 | philippej | using this : http://vserver.13thfloor.at/Stuff/AXIOM/it8-tungsten-01-12ms-perfect.png
| |
14:37 | philippej | hrbert posted it an hour ago
| |
14:37 | gcolburn | that's the one I used :) except I have to convert to TIFF
| |
14:38 | philippej | me too
| |
14:38 | gcolburn | what commands are you using? I'll see if I can reproduce it
| |
14:38 | philippej | scanin -v it8-tungsten-01-12ms-perfect.tif It8.cht R100604.txt
| |
14:38 | philippej | (R100604.txt file from Sebastian)
| |
14:39 | gcolburn | yeah. we should have the same one downloaded from the website
| |
14:39 | philippej | colprof -v -D"Apertus axiom qm R" -qm -R it8-tungsten-01-12ms-perfect
| |
14:39 | gcolburn | that command isn't any different than mine, except I've sometimes turned on perspective compensation
| |
14:40 | gcolburn | okay. you're colprof command is slightly different
| |
14:41 | philippej | I removed -a(something) to use the default
| |
14:41 | gcolburn | I get that lower error
| |
14:41 | gcolburn | let me see the difference in our commands
| |
14:41 | philippej | and -R doesn't change a lot, except I suspect photoshop likes it better this way
| |
14:42 | philippej | I followed the tutorial, and am quite newbie in this, so take with a grain of salt :-)
| |
14:42 | gcolburn | yeah. I was using a command from another tutorial
| |
14:42 | gcolburn | for profiling
| |
14:42 | gcolburn | but yours gives better results :)
| |
14:42 | philippej | documentation here : http://www.argyllcms.com/doc/Scenarios.html#PS4
| |
14:45 | gcolburn | there's more matrix outputs in that command then I'm used to with different gammas
| |
14:47 | philippej | icm file here : http://public.philippejadin.be/it8-tungsten-01-12ms-perfect%20qm%20r.icm can be installed and applied in photoshop for example to test
| |
14:47 | gcolburn | I think you need -am. that tells it to create a matrix
| |
14:47 | gcolburn | instead of a LUT for example
| |
14:48 | philippej | then error level is much higer
| |
14:49 | philippej | Profile check complete, peak err = 44.376942, avg err = 14.835942
| |
14:50 | gcolburn | yeah. that's more like what I was getting
| |
14:51 | philippej | but without am it's visually better
| |
14:51 | gcolburn | I would think a LUT could handle a wider range of profiling than a matrix, but with less resolution
| |
14:52 | philippej | as they say about am
| |
14:52 | philippej | If profiling a camera in RAW mode, then there may be some advantage in creating a pure matrix only profile, in which it is assumed that the camera response is completely linear. This may reduce extrapolation artefacts.
| |
14:52 | philippej | but
| |
14:52 | philippej | is the sensor perfectly lilear?
| |
14:53 | philippej | an of the shelf camera in raw mode is probably, but since we are a step before...
| |
14:54 | gcolburn | from what I saw of Herberts tests its pretty linear
| |
14:54 | gcolburn | but might be worth reviewing again
| |
14:55 | philippej | if I open the different icm files in icc profile inspector, I see that the files without am have some curves
| |
14:55 | gcolburn | yeah
| |
14:55 | gcolburn | the embedded LUTs presumably
| |
14:56 | philippej | I guess
| |
14:56 | philippej | do you want a screenshot?
| |
14:56 | gcolburn | DNG has tags to embed a matrix for the raw converter to use
| |
14:56 | gcolburn | ICC profiles most often use a LUT
| |
14:56 | gcolburn | sure
| |
14:58 | philippej | http://public.philippejadin.be/curves.png
| |
15:01 | gcolburn | interesting
| |
15:03 | philippej | and there are also info for white and black point
| |
15:03 | philippej | the tool is here : http://www.color.org/profileinspector.xalter (windows only)
| |
15:08 | S3bastian | time to leave the office, I took 2 halogen tungsten softboxes with me and will create new chart captures tomorrow with them
| |
15:08 | S3bastian | see you
| |
15:08 | S3bastian | left the channel | |
15:10 | philippej | see you !
| |
15:45 | philippej | left the channel | |
16:35 | gcolburn | left the channel | |
17:19 | gcolburn | joined the channel | |
17:19 | Bertl | wb gcolburn!
| |
17:55 | gcolburn | left the channel | |
18:32 | se6astian | joined the channel | |
18:40 | Bertl | wb se6astian!
| |
18:53 | se6astian | thanks :)
| |
18:54 | se6astian | Jay replied, classic misunderstanding, good this is solved
| |
18:56 | Bertl | excellent
| |
18:57 | jucar | joined the channel | |
18:59 | se6astian | and I took two tungsten softbox lights home from the university and took a day off tomorrow
| |
19:00 | se6astian | so I will shoot new charts tomorrow hopefully with fully uniform lighting and 100% accurate color temperature
| |
19:00 | Bertl | those are tungsten or halogen?
| |
19:00 | Bertl | because halogen != tungsten (lightwise)
| |
19:04 | se6astian | seems to be halogen
| |
19:04 | se6astian | http://www.bhphotovideo.com/c/product/285875-REG/Lowel_LC_95LBZ_Rifa_Lite_EX55_Softbox_Light.html
| |
19:06 | Bertl | okay, the spectrum is good on halogen AFAICT, but it has a different color
| |
19:15 | Bertl | btw, I created another perfect chart from my early daylight scene
| |
19:15 | Bertl | http://vserver.13thfloor.at/Stuff/AXIOM/scene_daylight_211ms_c2-perfect.png
| |
19:15 | Bertl | might be interesting to evaluate that one
| |
19:17 | se6astian | thank, I will try this one in Lprof now
| |
19:22 | se6astian | Average dE :
| |
19:22 | se6astian | 11.09
| |
19:22 | se6astian | Standard deviation :
| |
19:22 | se6astian | 8.35
| |
19:22 | se6astian | Peak :
| |
19:22 | se6astian | 57.23
| |
19:22 | se6astian | Min :
| |
19:22 | se6astian | 0.80
| |
19:22 | se6astian | Average Target error: (99% confidence)
| |
19:22 | se6astian | the brighter chart had an average dE (I guess this is the error) of around 6
| |
19:22 | se6astian | so as troy suggested slightly overexposed images seem to be better for profiling
| |
19:24 | Bertl | and average dE means?
| |
19:24 | se6astian | "deviation error" I guess, trying to find references on the net atm
| |
19:25 | Bertl | I think it is essential to put not just a name to those values but also a unit and an explanation what is actually measured
| |
19:25 | Bertl | otherwise it is just random numbers with a hope that smaller means better
| |
19:26 | Bertl | for example, the std deviation, 8.35, what deviates from what here?
| |
19:26 | Bertl | for sure, the colors within each tile do not deviate at all
| |
19:26 | se6astian | dE is short for delta E which is a standard way to measure color errors.
| |
19:26 | se6astian | http://lprof.sourceforge.net/help/checker.html
| |
19:27 | Bertl | what does peak 57.23 mean? what does peak here, where and why?
| |
19:28 | Bertl | and who is writing such a help with so many typose/grammatic errors :)
| |
19:28 | Bertl | shows how close the the color will be once the profile is applies
| |
19:30 | Bertl | so dE of 11.09 basically means (when the decibel part is correct) that the color is off by a factor of 10 or more?
| |
19:31 | se6astian | lprof profile inspection tab is very interesting
| |
19:31 | se6astian | it shows each color seperately
| |
19:31 | se6astian | what the color should be, what it was
| |
19:31 | se6astian | error in hue, chroma, luma
| |
19:32 | Bertl | so the error is in comparison to a perfect chart then?
| |
19:32 | Bertl | i.e. difference of each color to the assumed chart color?
| |
19:32 | se6astian | could be, not entirely sure yet
| |
19:32 | se6astian | would not make much sense tbh though
| |
19:33 | Bertl | if so, we will never get a lower dE, on the contrary, it will always be wrong
| |
19:33 | Bertl | but that would explain why the DSLR scores way better
| |
19:34 | Bertl | i.e. their 'raw' image will already be post processed regarding color calibration, at least to some degree
| |
19:34 | se6astian | This report shows how close the color will be once the profile is applied to the image.
| |
19:34 | se6astian | so it does take the created profile into account
| |
19:35 | Bertl | okay, so why is it off by a factor of 10 ?
| |
19:35 | Bertl | and where is it off by a factor of 10?
| |
19:40 | se6astian | good questions, no idea :)
| |
19:42 | Bertl | btw, I'm sure there is a list of color values used for the original chart
| |
19:42 | Bertl | it would be trivial to create a chart with those color values if you can put them in a list
| |
19:43 | se6astian | sure, but what would we do with that list?
| |
19:50 | se6astian | off to tv :)
| |
19:58 | Bertl | create a perfect chart and check what the various tools produce?
| |
20:01 | gcolburn | joined the channel | |
20:04 | Bertl | wb gcolburn!
| |
20:04 | gcolburn | hello!
| |
20:05 | Bertl | http://vserver.13thfloor.at/Stuff/AXIOM/scene_daylight_211ms_c2-perfect.png
| |
20:06 | gcolburn | ah daylight
| |
20:07 | gcolburn | not sure if you saw this conversation earlier, but when philippej ran Argyll using a LUT instead of a matrix he had errors less that 4%
| |
20:07 | gcolburn | I was able to reproduce it
| |
20:08 | gcolburn | I've sort of wondered myself when you run into limitations of a 3x3 matrix… but most literature I read indicates a matrix is best for a camera profile
| |
20:10 | Bertl | well, we can do both as you know :)
| |
20:11 | Bertl | or to be precise, we are somewhere in between a 3D lut and a matrix
| |
20:11 | gcolburn | yeah
| |
20:11 | gcolburn | I got 14% error with that one
| |
20:11 | Bertl | do we already know what that means?
| |
20:12 | gcolburn | "If the -v flag is used (verbose), then at the end of creating a profile, the maximum and average fit error of the input points to the resulting profile will be reported. This is a good guide as to whether things have gone smoothly in creating a profile. Depending on the type of device, and the consistency of the readings, average errors of 5 or less, and maximum errors of 15 or less would normally be expected. If errors are grossly higher than this, th
| |
20:12 | gcolburn | this is an indication that something is seriously wrong with the device measurement, or profile creation."
| |
20:12 | gcolburn | that's all the documentation states
| |
20:13 | Bertl | okay, so it basically calculates how good the transformed image matches the expected
| |
20:13 | gcolburn | yeah, basically what I figured it did
| |
20:13 | Bertl | and the value you are stating is the maximum error?
| |
20:13 | Bertl | or the average?
| |
20:13 | gcolburn | average
| |
20:13 | Bertl | what's the maximum?
| |
20:14 | gcolburn | 47%
| |
20:14 | Bertl | well, that is excessive
| |
20:14 | gcolburn | haha. yeah
| |
20:14 | gcolburn | but as I've said I haven't seen a visual difference between 12% and 20%
| |
20:14 | Bertl | can you tell which color causes this and is that for matrix or lut setup?
| |
20:14 | gcolburn | phillippej said that when he used the icc profile that was LUT based it looked better to him
| |
20:15 | gcolburn | I've been always testing matrix, except when I tried to reproduce the low errors with a LUT
| |
20:15 | gcolburn | if we wanted to switch to the LUT, we'd have to learn about the many LUTs that get embedded in the ICC profile and how to extract/use them
| |
20:15 | Bertl | okay, do we know what format the calibration files are?
| |
20:15 | gcolburn | they are icc profiles
| |
20:15 | Bertl | i.e. the ones which come with the chart
| |
20:15 | gcolburn | oh
| |
20:16 | Bertl | do you use them?
| |
20:16 | Bertl | or does it try to fit to an ideal chart?
| |
20:16 | gcolburn | there are 3 components
| |
20:17 | gcolburn | 1) Your image
| |
20:17 | gcolburn | 2) it8.cht which contains the edges I believe and the "ideal" values of each patch
| |
20:17 | gcolburn | e.g. the reference chart
| |
20:18 | Bertl | okay, can you point me to that file and do you know the format?
| |
20:18 | gcolburn | 3) the measurements of the individual chart that we have (R100604.txt)
| |
20:18 | gcolburn | its all ascii
| |
20:18 | Bertl | good
| |
20:18 | gcolburn | very human readable
| |
20:18 | Bertl | not so good :)
| |
20:19 | Bertl | the .cht is part of the argyll package, yes?
| |
20:20 | gcolburn | first you run "scan in". it recognizes the patches in your image and determines the values in your image. this outputs a .ti3 file (human readable with the results), then you run "colprof" which will try to fit the data with a matrix or LUT and create an ICC profile
| |
20:20 | gcolburn | yes
| |
20:20 | gcolburn | comes with argyll
| |
20:21 | gcolburn | in the .ti3 file there are XYZ, RGB mean values for each patch, and a standard deviation
| |
20:21 | gcolburn | so you can see which patches it had more variance in
| |
20:21 | Bertl | hmm, first question regarding it8.cht
| |
20:22 | Bertl | it says 290 boxes, 288 I can explain (12*22 + 24) what are the two missing ones?
| |
20:22 | gcolburn | not sure
| |
20:23 | Bertl | hmm, maybe boxes does not necessary mean color areas
| |
20:23 | Bertl | it looks like the color values are only 288
| |
20:24 | gcolburn | ok
| |
20:24 | gcolburn | when you run scan in you can pass it options to draw on various parts of the chart it detected and dump a diagnostic image
| |
20:25 | gcolburn | here's a semi-theoretical question. how good is 9 degrees of freedom for characterizing the response of an RGB sensor?
| |
20:26 | gcolburn | with a LUT, you have more degrees of freedom in principle
| |
20:26 | gcolburn | DNG uses a matrix
| |
20:29 | Bertl | LAB is basically XYZ just with different coordinate system, yes?
| |
20:31 | gcolburn | LAB is designed to be more perceptually uniform than XYZ. but it is a transform to go between the two
| |
20:31 | gcolburn | slightly more complicated than just a matrix transform
| |
20:31 | gcolburn | DNG requiries XYZ
| |
20:31 | gcolburn | but, we are not limited to that for the HDMI LUT implementation
| |
20:32 | gcolburn | in the LUT code I uploaded I go from camera to sRGB in one step, by combining the camera to XYZ and XYZ to sRGB transforms in one matrix (plus gamma encoding at the end)
| |
20:32 | Bertl | just looking at the R100604.txt and trying to make heads or tails out of it
| |
20:34 | Bertl | can we remove the R100604.txt easily from the equation?
| |
20:36 | Bertl | i.e. can we pretend that our IT8 chart is a perfect replica of the ideal chart?
| |
20:36 | gcolburn | well IT8 charts can be made for different purposes. I believe the R100604.txt is needed for Argyll to understand what to expect from the chart based on the photometric readings that Faust did on the chart
| |
20:36 | gcolburn | not sure
| |
20:36 | gcolburn | I think the issue is that you can't manufacture a perfect chart
| |
20:36 | gcolburn | and capture
| |
20:36 | Bertl | of course, I'm just asking because I would like to do some tests
| |
20:37 | Bertl | and eliminating the R100604.txt from the equation simplifies things
| |
20:37 | gcolburn | well there's nothing stopping us from trying our own calibration techniques
| |
20:37 | Bertl | the question now basically is, can we drop the R100604.txt or do we need to replace it by a dummy
| |
20:38 | gcolburn | we should read up on two things. 1) how the R100604.txt was created
| |
20:38 | gcolburn | 2) how scan in uses it
| |
20:38 | gcolburn | http://www.argyllcms.com/doc/scanin.html
| |
20:38 | Bertl | agreed
| |
20:45 | gcolburn | I think the XYZ values of the chart patches have to be measured for each chart as the material, ink, etc. may defer depending on manufacturer and intent of the chart (scanner, camera, printer, etc.)
| |
20:46 | Bertl | yes, thanks
| |
21:06 | gcolburn | have you seen this? http://www.argyllcms.com/doc/Scenarios.html#PS1
| |
21:06 | Bertl | soo, just to understand the chart (IT8)
| |
21:06 | se6astian | on my way to bed
| |
21:07 | Bertl | we agree that GS00 is white, yes?
| |
21:07 | se6astian | will take new captures of the it8 tomorrow
| |
21:07 | se6astian | please send me an email if you need any particular shots
| |
21:07 | Bertl | i.e. the first monochrome block at the bottom
| |
21:07 | se6astian | as I need to return the lamps on friday
| |
21:07 | Bertl | well, try to make as many snaps as possible
| |
21:07 | Bertl | and keep good documentation
| |
21:08 | se6astian | but with what variations? :)
| |
21:08 | se6astian | I will
| |
21:08 | Bertl | always do several shots without changing the position
| |
21:08 | se6astian | noted
| |
21:08 | se6astian | good night!
| |
21:08 | Bertl | so that we can remove the dynamic noise
| |
21:08 | se6astian | left the channel | |
21:08 | Bertl | gcolburn: do we agree that GS00 is white?
| |
21:09 | gcolburn | yeah
| |
21:10 | Bertl | okay, so when I take those values as XYZ (from the IT8.cht) and convert it to sRGB, what values can I expect?
| |
21:11 | Bertl | because the onlineconverter I found gives me #FBE8CC
| |
21:12 | gcolburn | well, assuming the conversion is correct then that's what you would get if the incident light from the patch really was the XYZ value from the chart
| |
21:13 | gcolburn | by the way, I read this:
| |
21:13 | gcolburn | "Well behaved input devices will probably give the best results with a shaper/matrix profile, and this may also be the best choice if your test chart has a small or unevenly distributed set of test patchs (ie. the IT8.7.2). If a shaper/matrix profile is a poor fit, consider using a LUT type profile."
| |
21:13 | Bertl | so the chart shouldn't be white when it would be perfect or how to interpret this?
| |
21:14 | Bertl | I would have expected the XYZ 'white' to result in (s)RGB white or at least a gray, no?
| |
21:15 | Bertl | the assumed 'black' also gives
| |
21:15 | Bertl | #0E090A which is a dark brown
| |
21:17 | gcolburn | hmm
| |
21:17 | gcolburn | it also depends on your monitor being calibrated
| |
21:18 | gcolburn | I finally got my colorimeter to work a month ago on my computer and got my NEC LCD profiled. the channels were definitely off :)
| |
21:19 | gcolburn | but as far as the numerical values go, I would initially think so it would be 255, 255, 255 for example but I'm not entirely sure
| |
21:20 | gcolburn | I've noticed when I view the DNG there is a brownish ting to the grayscale patches
| |
21:21 | gcolburn | kind of brown. i've always thought it was due to the fixed pattern noise. so not sure
| |
21:29 | gcolburn | here's what the Argyll docs say about the .chrt file
| |
21:29 | gcolburn | "The XYZ values are assumed to be scaled to a maximum Y value of 100. An expected color value doesn't have to be provided for every defined sample box, nor is it expected to be accurate - it just has to represent the approximate expected color. (Actual chart reference values are provided as a separate CGATS file to scanin)."
| |
21:31 | jucar | left the channel | |
21:38 | Bertl | I think we can agree that #0E090A is not black regardless how my monitor is calibrated (as it is not involved :)
| |
21:39 | Bertl | anyway, just trying to figure out what the it8.cht values actually mean
| |
21:42 | gcolburn | yeah
| |
21:42 | gcolburn | well according to that documentation, they don't have to be accurate
| |
21:43 | gcolburn | as its read in from another file
| |
21:43 | gcolburn | which I presume is the measured from the manufacturer (trying to verify that)
| |
21:44 | Bertl | so where does the CGATS file come from?
| |
21:44 | dmj_nova | joined the channel | |
21:44 | gcolburn | its a more general format not specific to Argyll
| |
21:44 | gcolburn | I'm pretty sure its the R100604.txt file
| |
21:44 | gcolburn | but I want to verify that
| |
21:45 | Bertl | so the it8.cht is not used at all for the calibration? except for where to place what color box?
| |
21:45 | gcolburn | http://www.colorwiki.com/wiki/CGATS.17_Text_File_Format
| |
21:45 | gcolburn | yes, what you said is my understanding
| |
21:46 | gcolburn | the file from faust is in the CGATS format
| |
21:46 | gcolburn | so for coming up with your own calibration
| |
21:46 | gcolburn | take the values from that file
| |
21:47 | gcolburn | so GS0 is 76.53 79.65 68.26
| |
21:48 | Bertl | which is more reddish than the beige we had before, but otherwise definitely not white
| |
21:51 | Bertl | unless I'm completely wrong, white should map to ~95/100/109 in XYZ
| |
21:51 | guesst | so where is the issue? why do gray references appear non gray ?
| |
21:51 | gcolburn | where do you get that from?
| |
21:51 | Bertl | from the transformation matrix
| |
21:52 | gcolburn | with what input value?
| |
21:52 | Bertl | #FFFFFF as RGB
| |
21:53 | Bertl | guesst: the issue/question is, why would the IT8 chart purposely use non-gray values for gray fields?
| |
21:53 | Bertl | (i.e. colors in the gray section)
| |
21:54 | gcolburn | my guess is you can't manufacture a perfect gray
| |
21:54 | Bertl | in this case the white? hello?
| |
21:55 | Bertl | I would say, the white photo paper is pretty white, no?
| |
21:55 | Bertl | and not red or orange
| |
21:56 | Bertl | but okay, let's assume the printed IT8 chart has orange-beige in the white field
| |
21:56 | gcolburn | yeah, but to what accuracy/precision?
| |
21:56 | Bertl | there must be some definition what color it actually should be
| |
21:56 | gcolburn | that's why paper profiles are important
| |
21:56 | gcolburn | yeah
| |
21:57 | Bertl | i.e. it should be 'white'
| |
21:57 | gcolburn | its in the IT8 standard
| |
21:57 | gcolburn | i'm trying to find a copy
| |
21:57 | gcolburn | so we can read it
| |
21:57 | Bertl | but that value is not represented in the it8.cht for whatever reason, yes?
| |
21:57 | Bertl | i.e. they put some color there which they know is different from what it should be, do I get that right?
| |
21:58 | Bertl | sounds a little arbitrary to me
| |
21:58 | gcolburn | I have no idea how they designed the chart. but obviously the goal is to cover a wide gamut of colors
| |
21:58 | Bertl | I can accept that the it8.cht color data is just to show achart
| |
21:59 | Bertl | i.e. visualize the chart, but why would you put 'wrong' colors there?
| |
21:59 | gcolburn | I have no idea. I thought that was odd
| |
22:03 | gcolburn | color science is very complicated. some day when I have the time I'd like to read a good book on it
| |
22:04 | Bertl | I think it is made unneccessarily complicated by a multitude of (half)standard and accidents
| |
22:05 | gcolburn | yeah. I think its that way in many fields though
| |
22:05 | gcolburn | for example in radiation therapy most aspects of "dosimetry" are defined by standards
| |
22:05 | gcolburn | which change over time
| |
22:06 | gcolburn | usually trying to account for more things that had to be neglected previously
| |
22:06 | gcolburn | I'd like to try running shots from various cameras through Argyll and see what we observe
| |
22:18 | Bertl | well, I don't think that we will get _any_ raw data from those cameras
| |
22:19 | gcolburn | yeah
| |
22:19 | Bertl | i.e. it will always be somewhat processed
| |
22:19 | Bertl | anyway, off to bed now ... have a good one everyone!
| |
22:19 | gcolburn | goodnight!
| |
22:43 | dmj_nova | left the channel | |
22:50 | gcolburn | left the channel |