| Forums | Polls | Register | Reply | Search | Statistics |
Display Calibration Light Illusion Forums / Display Calibration /   
 

New User Attempting Calibration of a RS4500

 
Author movieguy
ZRO
Male
#1 | Posted: 12 Oct 2020 21:57 
I don't know what possessed me to think learning to calibrate displays during COVID would be a relaxing new hobby, but here I am.

I am trying to calibrate my RS4500 using a 3D LUT and a MadVR envy. I have already spent a fair bit of time making mistakes and failing (but learning) some of the basics and have read a fair bit on the forums as well as the very useful user guides on this website. I will spare you the learning curve stories and cut to the latest work flow I have followed - still to a failed result.

RS4500 Pre-Setup
Colore Profile: DCI-No Filter (maximizing light output at the expense of color gamut)
Color Temp:6500K
Grayscale 2.2
Medium Laser

Klein Settings:
Intelligent Integration:Off
Integration Time: 4 sec
Extra Delay: .25 sec

I am using a Klein K10a for the measurements that has been matched to an i1Pro. I start by selecting the target gamut of DCI P3 D65 Gamma 2.2 and use manual 100% white to adjust the RGB gain settings in the projector to minimize the deltaE at 100%. I then take a Primary/Secondary run to identify initial issues. I am obviously not able to hit full DCI gamut without the filter in place so I can see errors in the red/green primaries and I see clipping near the end of the red scale (where multiple red tests resolve to a single point). I save the Prim/secondary run and extract the native colorspace of the projector. I use this as a subspace gamut with target code values selected. I do another prim/secondary run to make sure things look good and the clipping has been eliminated. I had some issues where activating the subspace was actually causing the gamut to shrink to rec709, but that issue has not appeared in my latest runs today. Everything SEEMS to look good up until this point. I go to the Characterization menu and select a 21^3 cube run. I enabled drift compensation and run the profile. It completes without event in about 2 hours 20 minutes.

I have tried both targeting the DCI colorspace with the resultant data using concatenation of LUTs as documented in the user guides as well as simply targeting REC709 which should work without concatenation given the projector's profile covers the 709 gamut. In either case the resultant LUT looks bad with what appear to be issues in the blue corner. I have uploaded these to the Envy (both rec709 and the DCI) and neither looks right. I Can tell the 709 profile is off because I would expect when I enable/disable it I should see a significant change is red output as the projector is set to DCI.

Its taken me several attempts to get this far and realize some errors I have made, but I am stuck at the moment and not sure what I am missing. I am sure this is another user error (or more!) on my part and could use some suggestions by someone with more experience. I have attached my bcs file from the latest failed run. Any input is greatly appreciated. If I left off any information, I am happy to answer questions. Thanks!

RS4500 DCIP3 Colorspace with.zip Attached file:
RS4500 DCIP3 Colorspace with.zip

 

Author jfinnie
ZRO
#2 | Posted: 12 Oct 2020 23:00 | Edited by: jfinnie 
I've had a few JVC projectors - unfortunately not up to the dizzy heights of the RS4500. Currently have an X7900.

Anyway, JVC's internal colour management is a bit "exciting". I think it probably works reasonably for consumer use without 3DLUTs (in fact it seems from hacking around with the Autocal network traffic it may actually be 3DLUT based internally in the projector!), but their display profiles have odd issues around gamut edges where the colour management is most evident, and those issues present an odd response that it seems LUT engines struggle with.

I would consider putting the projector into "profile off" mode and profiling it like that (without going near all the subspace etc stuff - so just a straightforward profile of the projector response to input stimulus) and see how that goes. The profiles in "profile off" - at least on my display - are completely regular and linear in nature.

It may be that blue - coming from the laser - is being adjusted by the colour management in a way which isn't conducive to a good LUT.

For what it is worth I had mixed experience with the process of reducing the gamut through the concatenation process; it seemed that the gamut you had to choose was quite a lot smaller that the native gamut of the display before all the edge artifacts would go away; so much so that I decided I didn't like the technique and think getting the projector into a more linear place - like using the Profile Off function - was a better option.

Your drift chart is interesting too, the pattern is a bit unusual. Though I don't think it is the cause of the issues, it merits understanding. I mean if you believe the chart the colour temp is changing significantly during the profile in quite a distinct fashion. On OLEDs this kind of thing might be attributed to the patch sequence itself making the panel drift, but I've not seen this on D-ILA before.

Are you hearing the fans ramping up and down at all while profiling it? I wonder if alternating cooling / heating of the laser is causing this, or something else.
If you run the same patch sequence, in the same order, do you always get the same drift pattern?
Just checking - definitely got dynamic laser dimming disabled?

Sometimes funny drift patterns can be a symptom of a patch delay that isn't long enough and so you are seeing some bleed through of the previous patch into the white patch (because it is measuring part of the previous patch and part of the new patch) - it is worth checking that this isn't happening.

Anyway, they're just some thoughts, hopefully useful to you. Let us know how you get on.

Author movieguy
ZRO
Male
#3 | Posted: 12 Oct 2020 23:47 
That you so much for the input and thoughts! I may try a profile completely off run and see if I can work with that. I did that early on and had some issues with posterization in the rec709 LUT when viewing material. Now that was something I tried several runs ago and my approach has been pretty iterative, so its possible that there was something else in my technique early on that was causing those issues. I also found that my contrast was reduced in that mode, which may just require me to spend more time with the pre-calibration configuration of the display. It is certainly worth trying again.

It is interesting to note that the red and green (where the gamut isn't really covered well looks better than the blue (where I supposedly have good coverage). I am wondering if maybe I need to rein blue in a bit if I am going to try the subspace method gain. Maybe its not quite linear near the edge and is causing the issues.

I was also curious about my drift chart. I think I have seen a similar pattern in the last run I made as well. I should have kept all my failed profiles to compare, but I deleted them after realizing the output wasn't really working for LUT generation. I may try another run tomorrow and see if I can replicate the drift pattern. I am not hearing the fans change speeds, but to be honest I have started using TeamViewer to monitor the calibration remotely so I don't have to be in the dark room for 2+ hours. Part of my process is to turn the temp down in the room to 68 Fahrenheit and I'm able to keep it there for the duration of the calibration - this has seemed to really help keep the projector steady. I definitely have the laser dimming disabled as well. I appreciate your asking. I keep hoping its a setting like that, which I have overlooked.

I was also thinking of increasing the additional time for under the probe settings to see if I had that setting too short and the previous patch was bleeding into the results - as you suggested. Now that I think about it I believe the delay on the JVC is pretty long (like 500 ms when setting a lip sync setting). I believe that might fit because the extra delay I believe is from the time the patch generator reports back it is displaying the requested patch. I will certainly try that tomorrow and see where that gets me. I may try it with a reduced sized cube just to make sure before taking on the 21^3 monster again.

Thank you again for the input. You have certainly given me some things to try and experiment with. As I progress I will update on any discoveries.

Author jfinnie
ZRO
#4 | Posted: 13 Oct 2020 00:42 | Edited by: jfinnie 
Contrast changing for the worse from using profile off would seem pretty unusual, but bear in mind I don't have RS4500 experience.
It should have the most contrast possible for a given optical setup (laser level, dimming, iris, etc), because it is supposed to bypass all the CMS. In real life all the CMS controls can actually do with respect to a genuine 100% input stimulus is apply cuts to the maximum panel capability (edit; should have said 100% white stimulus. For 100% colours, it can cut the primary and / or add other primaries to change the hue or saturation). Of course, if you aren't at 100% input (for example, projector set to enhanced 0-255 or superwhite 16-255 mode, but you're only using video level patterns up to 235) then you could have different results, which would be related to what the gamma curve was doing for each mode at whatever level your "100%" test stimulus is.
I don't use MadVR, so I don't know about any funnies with that and the levels needed (I use a Lumagen that only has a video levels LUT).
I don't think with a projector there is any value calibrating to leave 235-255 on the table, as really there shouldn't be significant content there, and calibrating for it is really just giving away contrast.
My go-to setup for LUT gen is:
Profile off
Standard input mode
On X7900, contrast and brightness and colour all to 0. If you have ambient light polution or the RS4500 behaves differently this might not be right.
Note on the X7900 this can clip code 17, but I use my Lumagen black level control to tweak it back. As soon as you raise the brightness on the projector - although it brings code 17 back you're instantly giving away contrast because it just raises the black floor.

Colour temp high bright (this I think will set automatically with profile off).
At this point if you make sure the colour temp sliders (gains and offsets) are all 0 you should be able to measure max possible contrast and the maximum gamut possible for your projector (for a given cinema filter position). White point will be decidedly non-D65. The gamut should be linear. The gamma on the X7900 is pretty wild in profile off because all the autocal / factory cal gamma tables are bypassed. I don't know if RS4500 also behaves that way.
I'd then adjust the gains only on the colour temp control to bring a 100% patch into D65. You should only have to reduce max 2 sliders to get to D65 whitepoint. If you reduce 3 you will be giving away contrast. This should be your best possible apples to apples contrast measurement possible.
At that point do the cube profile. You don't need to (unless you're just curious) do the primary / secondary profiles etc.

Anyway, above works for me. Good luck!

Author movieguy
ZRO
Male
#5 | Posted: 15 Oct 2020 17:36 
Thanks so much for the great feedback, jfinnie. I learned a lot from how you laid out what you did any why. I was able to get what appears to be a good run in yesterday using your suggested failsafe settings - color profile off, hi-bright, and adjusting to d65 at 100% white (which turned out to be -79 green gain Red and blue gain 0). I also used a extra delay setting of .5 seconds (as opposed to .25 seconds in the previous run) to see if that impacted the measurements. Post calibration results looked really good for grayscale (under dE .4) and using a sampling of colors I was well under 1 dE. This worked for both a rec709 setting and a DCIP3 setting (obviously my primaries were higher dE because I cant cover the gamut for green and red). This was significantly better than the JVC autocal results by the numbers. I still need to watch some more material to assure I don't have any issues with posterization, but initial results look very promising! It was quite a thrill getting a good result!

I suppose I could stop there, but I am trying to understand what I am doing as opposed to just following a process, so I do have a few more questions.

For JVC projector calibrators - do you use stabilization in the settings? If so, what setting do you use? I do not use it, but I am curious if it is recommended for D-ILA JVC projectors. I ask because I am still curious about the drift chart in my profile run. I suspect something is still not quite right with the drift chart as the swings early on in my measurements (when the projector is switching between very low luminance points and very high luminance points). It usually seems to settle in later in the calibration, but the pattern usually holds, even when doing a run after the projector has been on for a couple of hours already.

When I profiled the projector using color profile off, I would have expected when comparing the results to the extracted colorspace of the profile that all the dots would have been green. I actually had a significant number of red dots. I thought this meant that the projector was not responding linearly to the inputs and I would have a bad LUT. Is this correct? I have attached the profile run for reference. Does this indicate I probably made some additional errors in the data collection phase?

Also, for the DCI or Rec2020 LUT I generate from this run, should I concatenate LUTs for this? I did not do that, but is it advisable because the colorspace of the projector is smaller than the target space?

For concatenation my understanding is that the process is: LUT1: source is the target colorspace (e.g. DCIP3) and the destination is the extracted colorspace of the projector (from the profile run). LUT2: source is the extracted colorspace of the projector and the destination is the profile run of the projector (17^3 or 21^3, etc). You then save LUT2 and go back to LUT1 and add LUT2 using the add menu of LUT1. Is this correct?

For comparison, when I was using the subspace calibration with a color profile on comparing the results to the extracted colorspace produced almost all green dots, with some yellows. I would have expected this to be a "good result", yet the LUTs generated from these were horrible. I suspect this issue was that I need to pull blue in a little more from the setting I used for the colorspace as the results do not look right in the blue primary with lower luminance at 100% blue going outside the 100% result. I think that was one of my issues. Does this sound correct? I've attached that profile run result as well for reference.

If you use a colorsubspace for calibrating to a set colorspace (e.g. DCIP3 or Rec2020), are the results only valid for generating a LUT to the target colorspace because you have "scaled" the RGB values to fit into what is the non-clipped colorspace the projector is capable of? I ask because when I generated a rec709 LUT directly from my profile run I did for Rec2020 with subspace active I found the resulting LUT didn't pull back the red to the rec709 CS properly. Did I miss a step somewhere when using a subspace? I was wondering how the LUT generation knows that that the measurements have been scaled to a smaller color gamut.

Finally, I do intend to keep at it and practice on my projector as I feel I am learning a lot and SLOWLY getting better at calibrating. Do I run the risk of damaging the display at all if I keep at it (as opposed to just watching source material)? While I was getting the hang of things I found myself doing a couple of characterizations a day with total time of about 6 hours (including setup, warm up, and manual measurements). I had done this for 3-4 days. I'm taking a break for a bit, but just want to make sure playing with colourspace on my display for extended periods doesn't cause issues.

To anyone who made it this far, thanks for reading! Sorry for the long post with a several questions, but I figured keeping all my questions in a single thread would be preferable. If you were reading this to learn how to calibrate your JVC I would say follow jfinnie's advice. Color profile off is the easiest way to get to what you are trying to accomplish.

Profile Runs.zip Attached file:
Profile Runs.zip

 

Author jfinnie
ZRO
#6 | Posted: 16 Oct 2020 22:04 
While your tenacity with the LUT concatenation process is to be admired, you're really wasting your time a bit if you can run with the profile off option

If you look at the two most recent profiles; first the profile off one, you can see your blue goes WAAAAY outside of REC2020 in the y direction (y is too small) and the hue is rotated (x is too large). In order to try to tame that back within 2020 (which the internal 2020 profile is doing) the projector has to add green to the blue to bring it back in line for y, but it can't do anything to reduce x, so when it gets to a certain extent all it can do is crop the hue which stays stationary. If you open the 2020 profile in Lightspace and enter a custom filter of "B == 1 && G == 0" you can examine the values of the points along the blue -> magenta edge of the gamut, and you will see as you get into the blue corner the values start stacking up on top of each other - basically for different input values you are getting the same output value. Similarly on the green -> yellow edge, as you get towards green the values stack on top of each other. You can look at the values using the custom filter "B==0 && G == 1". There is probably similar on red -> yellow and / or red -> magenta, but I gave up looking at this point! This causes the LUT engine issues.

In order to overcome those issues, you need to define a sub-gamut that doesn't include any of the duplicate points - so it needs to be using the x,y co-ordinates of a point inwards where there is no longer going to be multiple input values leading to the same output value.
In my experience it is pretty hard to do this. You can't assume the behaviour is the same at all stimulus levels, so the gamut has to be really quite small before the raggedy edges of the resulting LUT are truly removed.

Now, if you look at your profile off profile, it's basically perfect (minus the drift) in that you just have a pretty regularly spaced set of readings in the native gamut of the projector, all the way to the gamut edges.

You won't be able to make this profile off profile "go green" in the points viewer by changing the reference gamut to be the extracted gamut because:
1) the gamma of your profile will be nothing like the gamma in the extracted gamut. The gamut extraction assumes a power law gamma of the average of the profile gamma. You can see from your diff gamma it is miles off (it would have to be a flat line).
2) your profile doesn't have a flat RGB balance (I think it would be fine for it to be off d65, but it needs to be flat).
The above doesn't matter, it's not important for the purposes here; the 3DLUT can address both points. What matters is getting a good, clean 3DLUT out, which clearly this profile off profile manages. Comparatively, the LUT out of the 2020 profile is a disaster.

The issue with the input gamut "foldback" at blue to magenta edge and green to yellow edge - which you can see clearly in the 2020 profile - for me would write off even trying to use the 2020 profile when you have a perfect option available in using profile off. I believe there are folk who do persevere with this, but IMHO it's really not worth it. Profile off is the right mode to be using, everything else is making work for yourself to get a worse result...

Author movieguy
ZRO
Male
#7 | Posted: 17 Oct 2020 03:52 
You are correct - as you have been from the start. It's a fools errand to try and get a better calibration using a sub gamut, at least I have learned that from this exercise. Since my last post I have found discussions on other forums where people have spent a lot of time trying to avoid the edge compression in pre-calibrated modes and struggling as I have. I have no intention of pursuing this for a calibration of the projector. I just watched a movie this evening and it was fantastic. No issues with posterization or banding. I was very pleased. I am sure I could spend several more hours trying to figure out how to reduce the gamut in the projector's rec2020 mode, but I would loose some gamut, brightness, and contrast in the reduction to avoid the edge compression. Not to mention the hours I would spend doing it. Reading your response I know I still have a lot to learn (I have to figure out the custom filter option now!), but I am really thrilled to have gotten an amazing result out of using Colourspace. I have done previous LUTs using other solutions, but I have had issues with them, which was what lead me to Colourspace. I have learned more using Colourspace because its flexibility also requires you to have little more understanding of the techniques you are applying. I have enjoyed that, but it has also got me asking questions and trying to understand the "why" of it. Ironically, the results I have gotten with the Colourspace calibrations have been better than the other software solutions by their own metrics (the Colourspace LUTs test better than then other packages LUTs using their own verification tests!).

Also, your explanation of why I had a lot of red dots in for the profile off when comparing to its native colorspace is super helpful. I need to stew on it and look at it more closely, but am going to go back and re-exaimine the profile and make sure I understand WHY. Thanks for explaining that because it was something I was puzzling over.

The only reason I was pressing about subspace calibration is that I invested in the software package and the hardware so I could learn and also when I purchase a new projector or TV I am able to adapt to other constraints (like not having access to a profile off setting). So I will put this to bed with one last question - for theory's sake. If you are using LUT concatenation, you concatenate from source to destination, correct? In other words, you start with the LUT closest to the source, then add the next LUT in the chain and so on (if there are more than 2). I have read the guides on the website, so I think I understand the process, but would like to clarify this point as the guides make clear the order is very important. As an example, if I wanted to concatenate 2 LUTs for the profile off run I did (because the projector gamut is smaller than the target gamut) I could do 2 LUTs where:

LUT1= source is the DCI target colorspace ,destination is the native gamut of the projector with profile off
LUT2=source is the native gamut of the projector with profile off, destination is the 21^3 characterization of the projector with the profile off

Once you have these, you would go to the Maths tab of LUT1 and add LUT2. Yes?

Thank you, jfinnie. I owe you an adult beverage! I've taken up a fair amount of your time with what I am sure are fairly trivial questions for someone who has an in depth knowledge of video calibration. You have been very patient. I can assure you it is much appreciated. If the one eyed man is king in the land of the blind, you are an emperor by my estimation. Cheers.

Author jfinnie
ZRO
#8 | Posted: 17 Oct 2020 08:19 
If you want to get the extracted gamut referenced view to look better, you could profile the projector without having adjusted D65. That should give you a more or less flat RGB balance. It won't be totally flat because the gamma of your projector panels I think will be slightly different between the 3 channels, which when you think about it causes colours to vary from where they should be, as it means the rate they go from 0% output to 100% output is different. It's like mixing paint(ish).

There are arguments for and against adjusting the PJ to D65 in its menus using the RGB gains. In an ideal world you probably wouldn't use any display controls, and just leave all adjustment to the 3D LUT. However, in the real world, a lot of LUT holders are only video legal levels 16-235 (I know my Lumagen is in this camp). What this means is their last control point is at 235,235,235, so you can control the top corner of video white if you imagine a colour cube. If your content does have any values above 235 in it (can be from encoding issues, colour conversion in players, specifically crafted test content, etc) then because 255,255,255 doesn't have any LUT control, what you'll have is perfect white up to 235, an from 235 to 255 the whites will change colour towards the native white point of the projector. If you look at a whites (contrast) test pattern from something like Ted's disk, the visual effect you'll get is that you have a colour shift in your white at the top end.

I'm not sure the custom filters are enabled in Colourspace yet - at least I don't think I could work out how to get them up. I opened up the profile in Lightspace to view, which isn't as nice as you can't zoom into the corner of the CIE chart to fully see the stacking of the points.

I do understand why you want to try and get that workflow to work better; I've discussed with Steve a couple times and it would be nice if the SW would deal with this kind of display profile a bit better automagically. To be honest I think the concatenation etc features themselves works perfectly fine for what it is -general purpose LUT manipulations designed for colour professionals (which I am not, I'm just a bit further down the calibration rabbit-hole than you !). It is the application of the feature to this issue of these profiles that can be hit or miss, as I don't think many folk understand it enough to realise it only works once you get rid of all those duplicates appearing in the profile, which needs a quite reduced intermediate gamut for some displays.

I can't honestly remember the exact detail of which order the concatenation works with... I did work it out once but has since been forgotten as I don't use it anymore past originally working out that the gamut needed reducing a lot! If you want to try and work it out easier during your test, maybe use an obviously different gamut as the intermediate instead of one that is very close in size to the display native (it will make it easier for you to understand how it works).

If you are interested to play with other aspects of the SW - there is another approach to this also outside of LUT concatenation, which is to doctor the patch set used; so you can pre-treat the patch set with a suitably smaller sub-gamut to remove the patch areas that are problematic from the profile, export the patch list, and then just add back to the patch list manually the far extents of the colour cube (so probably the primaries and secondaries, ie (255,0,0) , (0,255,0) , (0,0,255) etc. At the time I was investigating I actually thought this worked better.

A well-mannered display should have a non-colour managed mode for when you want to do 3DLUT, which is what the Profile off mode is in the JVC. Unfortunately up until now, JVC have not implemented a Profile off mode with the cinema filter in. They're just about to release an update to the current consumer series N5-NX9 that finally enables this option (yay!). Without it you have to jump through a hoop to force the filter in with a "diagnostic" IP command. I know that with the built in colour profiles the filter causes way too much loss of light for a lot of folk; it would be interesting to know if you use profile off and force the filter in with the IP command exactly how much light is lost. It might not be anywhere near as lossy in Profile off with a 3DLUT externally than in the built in modes.

You previously asked about damage to your PJ - I think it is unlikely. I've never seen any hint of image retention or similar on my X7900. I'd still never leave a single patch on screen for hours on end though, that's just asking for trouble. In terms of your drift graph - I see a few possibilities. Either there is some drifting response of the panels to the level of the image displayed, or there is an inherent instability in the JVC projector (maybe it always cycles the fans from slow to fast to keep the laser just right, which causes the waves) or maybe the environment the PJ is in causes it to have to cycle the fans more aggressively.

OLED TVs behave in the first way. The patch sequences in Lightspace and Colourspace aren't perfectly balanced in terms of the luminance throughout the sequence; it used to be quite bad (there are some posts on here somewhere) and now they're very good, but there are still some better results for drift that can be had with a specialist patch set. I did some work a while back to try and help friends with OLED TVs which culminated in a patch sequence sorter that more or less eliminated their drift patterns.

You might try some of the patch sets attached. These have been sorted with an algorithm to try and balance for WRGB pixel displays, but I think it should also work for RGB pixel displays. The patch sets are interesting in that they also have a larger set of values in the greyscale, primaries and secondaries, which can sometimes be useful. It will be interesting to see if you still get that characteristic drift with these sets. For each size there are two sets; one is a pre-roll set which is used for getting a panel up to temperature, and the other is a full patch set.

Have fun anyway.

SOFS7.zip Attached file:
SOFS7 patch sets for OLED

 

Display Calibration Light Illusion Forums / Display Calibration /
 New User Attempting Calibration of a RS4500

Your Reply Click this icon to move up to the quoted message

 

 (?)
Only registered users are allowed to post here. Please, enter your username/password details when posting a message, or register first.

 

 
Online now: Guests - 1
Members - 1 [ lucacasa ]
Max. ever online: 264 [6 Jun 2020 12:02]
Guests - 264 / Members - 0