Improving Madvr HDR to SDR mapping for projector - Page 270 - AVS Forum | Home Theater Discussions And Reviews
Forum Jump: 
 8015Likes
Reply
 
Thread Tools
post #8071 of 8375 Old 11-19-2019, 01:10 PM - Thread Starter
AVS Forum Special Member
 
Soulnight's Avatar
 
Join Date: Dec 2011
Location: Germany
Posts: 1,479
Mentioned: 226 Post(s)
Tagged: 0 Thread(s)
Quoted: 1130 Post(s)
Liked: 1858
Quote:
Originally Posted by madshi View Post
Thanks for testing! Could you post the numbers you used for L-80 and Mirror? I don't think anyone has posted Mirror numbers for the new madVR build yet, and since you designed the original Mirror curve, you're probably the best person to decide how to port Mirror to the new madVR build.
Here the different curves I have compared with L-80 still being my favorite.
(Linear interpolation was used for all the missing data points)
Log 50% strength or Latest Neo-XP 100% strength are very similar in my opinion --> good for a low strength setting

Log: Strength 50
1000-1150-1300-1450-1750-1974-2310-2640-2870-3050-3200-3330-3440-3620-3770-4000

L80: strength 50
800-860-920-980-1100-1220-1400-1700-2000-2300-2600-2900-3200-3800-4400-5000

Mirror: strength 50
250-250-250-250-250-350-500-750-1000-1250-1500-1750-2000-2500-3000-4000

Neo-XP / Manni Tweaked :
(strength 125): 1000-1050-1100-1150-1250-1350-1500-1750-2000-2000-2000-2000-2000-2000-2000-2000

Latest and greated Neo-XP: Pure Log, strength 100:
750-1019-1134-1202-1287-1343-1403-1471-1519-1556-1586-1612-1634-1672-1702-1750

Latest and greated Fer15: My new suggestion: (HSTM strength 35)
380 - 880 - 1280 - 1590 - 2070 - 2400 - 2760 - 3140 - 3380 - 3540 - 3660 - 3750 - 3820 - 3930 - 4000 - 4100
Soulnight is offline  
Sponsored Links
Advertisement
 
post #8072 of 8375 Old 11-19-2019, 01:40 PM
Member
 
Join Date: Nov 2017
Location: Stockholm, Sweden
Posts: 96
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 71 Post(s)
Liked: 23
Could you please share a screenshot, like the ones Manni shares, of your L80 (your desired curve).

Thank you!

Sent from my SM-N975F using Tapatalk
arcspin is offline  
post #8073 of 8375 Old 11-19-2019, 01:47 PM
Senior Member
 
Icaro's Avatar
 
Join Date: Jul 2007
Location: Italy
Posts: 418
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 140 Post(s)
Liked: 76
Quote:
Originally Posted by Soulnight View Post
Here the different curves I have compared with L-80 still being my favorite.
(Linear interpolation was used for all the missing data points)
Log 50% strength or Latest Neo-XP 100% strength are very similar in my opinion --> good for a low strength setting

Log: Strength 50
1000-1150-1300-1450-1750-1974-2310-2640-2870-3050-3200-3330-3440-3620-3770-4000

L80: strength 50
800-860-920-980-1100-1220-1400-1700-2000-2300-2600-2900-3200-3800-4400-5000

Mirror: strength 50
250-250-250-250-250-350-500-750-1000-1250-1500-1750-2000-2500-3000-4000

Neo-XP / Manni Tweaked :
(strength 125): 1000-1050-1100-1150-1250-1350-1500-1750-2000-2000-2000-2000-2000-2000-2000-2000

Latest and greated Neo-XP: Pure Log, strength 100:
750-1019-1134-1202-1287-1343-1403-1471-1519-1556-1586-1612-1634-1672-1702-1750

Latest and greated Fer15: My new suggestion: (HSTM strength 35)
380 - 880 - 1280 - 1590 - 2070 - 2400 - 2760 - 3140 - 3380 - 3540 - 3660 - 3750 - 3820 - 3930 - 4000 - 4100
Sorry Soulnight
but I think this is Manni's L-80 curve:
(strength 50) 80-110-140-170-200-230-260-290-320-380-440-500

Am I wrong, or are you referring to another curve?

Sorry for my bad English
Epson LS10500 - AMD RX580 - LightSpace CMS

Last edited by Icaro; 11-19-2019 at 02:08 PM.
Icaro is online now  
Sponsored Links
Advertisement
 
post #8074 of 8375 Old 11-19-2019, 01:51 PM
Member
 
Join Date: Oct 2019
Location: Netherlands
Posts: 29
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 7 Post(s)
Liked: 13
I Just completed my testing time..
For now i still prefer the L80/L63.
There is more depth in the picture with my 440JVC when i play with the L curves.
woofer, Manni01 and Soulnight like this.
Reezepees is offline  
post #8075 of 8375 Old 11-19-2019, 02:17 PM
Newbie
 
Join Date: Nov 2019
Posts: 9
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 1 Post(s)
Liked: 11
Quote:
Originally Posted by Soulnight View Post
Here the different curves I have compared with L-80 still being my favorite.
(Linear interpolation was used for all the missing data points)
Log 50% strength or Latest Neo-XP 100% strength are very similar in my opinion --> good for a low strength setting

Log: Strength 50
1000-1150-1300-1450-1750-1974-2310-2640-2870-3050-3200-3330-3440-3620-3770-4000

L80: strength 50
800-860-920-980-1100-1220-1400-1700-2000-2300-2600-2900-3200-3800-4400-5000

Mirror: strength 50
250-250-250-250-250-350-500-750-1000-1250-1500-1750-2000-2500-3000-4000

Neo-XP / Manni Tweaked :
(strength 125): 1000-1050-1100-1150-1250-1350-1500-1750-2000-2000-2000-2000-2000-2000-2000-2000

Latest and greated Neo-XP: Pure Log, strength 100:
750-1019-1134-1202-1287-1343-1403-1471-1519-1556-1586-1612-1634-1672-1702-1750

Latest and greated Fer15: My new suggestion: (HSTM strength 35)
380 - 880 - 1280 - 1590 - 2070 - 2400 - 2760 - 3140 - 3380 - 3540 - 3660 - 3750 - 3820 - 3930 - 4000 - 4100

Uh, thank you for posting this numbers again, as I had a completely other curve for Mirror and I don't know why. Well that did it for me again. That means also that I would take Mirror as my first choice.



Still I like Neo-XP's new one (Pure Log, strenght 100) and also 2.0. Close to that Manni's LLH-800.
The L80 seems ok for me, but not my choice.
Manni01 and Soulnight like this.

JVC RS1000 (N5) - 155'' AT-Screen - Marantz AV8805 - Klipsch LaScala (F/L) Cornwall (C) Heresy (BS) KS-525 THX (SideS) KW-120 THX (x4 - SBA); Genelec 8040 (Heights x6) - Oppo 203 - RTX2080 i5 8400 HTPC
Jan.2000 is offline  
post #8076 of 8375 Old 11-19-2019, 02:23 PM - Thread Starter
AVS Forum Special Member
 
Soulnight's Avatar
 
Join Date: Dec 2011
Location: Germany
Posts: 1,479
Mentioned: 226 Post(s)
Tagged: 0 Thread(s)
Quoted: 1130 Post(s)
Liked: 1858
Quote:
Originally Posted by Icaro View Post
Sorry Soulnight
but I think this is Manni's L-80 curve:
(strength 50) 80-110-140-170-200-230-260-290-320-380-440-500

Am I wrong, or are you referring to another curve?
With the beta 112b, as madshi explained, all numbers need to get multiplied by 10. Also there are a few more steps added by madshi in v112b which I have linearly interpolated to exactly match v110.

If you use the old HSTM numbers of v110 in v112b, the curve will be completely wrong and nothing like it was in v110.
stevenjw, woofer, Manni01 and 2 others like this.
Soulnight is offline  
post #8077 of 8375 Old 11-19-2019, 02:26 PM
Senior Member
 
Icaro's Avatar
 
Join Date: Jul 2007
Location: Italy
Posts: 418
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 140 Post(s)
Liked: 76
Ok Soulnight,
Thanks for the explanation.

Sorry for my bad English
Epson LS10500 - AMD RX580 - LightSpace CMS
Icaro is online now  
post #8078 of 8375 Old 11-19-2019, 03:27 PM
AVS Forum Special Member
 
Manni01's Avatar
 
Join Date: Sep 2008
Posts: 9,410
Mentioned: 366 Post(s)
Tagged: 0 Thread(s)
Quoted: 5690 Post(s)
Liked: 6026
Quote:
Originally Posted by Soulnight View Post
So, we just tested the latest Neo-Xp and Fer15 curve and also the previous Neo-Xp with 125 proposed by Manni01.

Against, old LOG, L-80 and Mirror.

Well... I really don't understand at all the enthusiasm about the new Neo-xp curve (or Fer15). It's basically identical to the old Log.

We checked multiple times if we did not do any mistakes, typos or had issues with profiles because new Neo-Xp and old Log from Neo-xp are virtually identical with our settings:
50 nits
75 dynamic tuning
No shadow recovery
Highlight Recovery: high

So Log or new Neo-xp can still for me act as a low strength HSTM.

In no way they can replace L-80 for me. It's not even close.
Mirror as always is like a stronger L-80 so can't be replace either.

I really am surprised and feel confused based on what I read here about those new curves and I really wanted to say: "yeah, that's the curve". And it ends to look identical to LOG which is maybe in the end not that surprising since both were designed by Neo-Xp.

Fer15 curve is slightly weaker than Log / New Neo-XP.
Manni01 modified version with the 125 strength is slightly stronger but still much weaker than L-80.

So it seems Neo-xp taste are still the same but for some other people, their taste changed or they did not directly compare with the old L-80??

I guess it would be good to give the new numbers for all the curve since some people may have difficulty to interpolate and convert L-80, Log or Mirror to the new format of beta v112b.
Well, as we've already discussed, you are clearly compressing a lot more with peak 50 / dt 75, so it might explain why you need a stronger curve.

Either L-80 doesn't exhibit the same issues with so much compression, or you don't care about them.

For me there are big differences between the older curves and the more recent ones.

With the older ones, I always felt that we were leaving a lot of contrast on the table (or they were too milky).

The issues I saw in L80 when adapting it to build 112 made me step away from it and try to find a better compromise (LLH-800, which you haven't tested).

The recent ones, with 110nits peak and 100 DT, get very close to LLH-800 contrast-wise, but deal much, much better with difficult content.

So I'm not denying the fact that they don't work for you, but for me it's just that the issues (flatness, milkiness or harsh skin tones) with LLH-800 or L-80 20% of the time simply override the minor contrast gain 80% of the time. I don't mind when a curve is a bit excessive, but I do mind when it's significantly worse or wrong in some situations (I have mentioned detailed frame numbers, so I assume that you have tested them and heven't seen any issues or don't mind the poorer performance in that case, if it behaves the same with your settings as here).

A good way to understand the differences would be to test using the settings I posted for the various curves. It would be too dark for you, but hopefully you'd be able to see what it's about and why it's a bid deal for those with more headroom for highlights than you have.

Looking forward to testing Neo-XP's V2.0 as I expect it will be a bit more contrasty based on the curve, but I might not have the time to do so tonight.

Batch Utility V4.02 May 16 2019 to automate measurements files for madVR with support for BD Folders
JVC Macro feature on Vertex/Vertex2/Integral2/Maestro
Manni01 is offline  
post #8079 of 8375 Old 11-19-2019, 04:15 PM - Thread Starter
AVS Forum Special Member
 
Soulnight's Avatar
 
Join Date: Dec 2011
Location: Germany
Posts: 1,479
Mentioned: 226 Post(s)
Tagged: 0 Thread(s)
Quoted: 1130 Post(s)
Liked: 1858
@Manni01

I don't have an issue at all that you like the new one better.
I am not trying to convince you. Just reporting what I saw.

I would prefer to see the same thing that you do and that we agree but it's also ok when it does not happen.

I personally did not see any advantage of the new log from Neo-xp over the old log. Like I did not see any advantage of L-80 over L-63 before that.
Maybe you can provide some examples where new log has better contrast than old log.

For me, it's more like 99% of the time L-80 has a much better contrast than any Log while still feeling natural. In very rare cases, log may looks better.

In any case, I am happy that we have (still) now some good curves for different taste/ setup.

1) old log / new log (the safe setting, no matter what).
2) L-80 ( strong contrast)
3) Mirror (maximum pop)

And I do like "log" by the way. Old and new. It's a safe setting and still much better than HSTM off. I am just discussing preferences at this point. And it's important for me to also have the log in the final HSTM settings so that if my taste/ setup evolves, I can roll to the pure log instead.

Last edited by Soulnight; 11-19-2019 at 04:42 PM.
Soulnight is offline  
post #8080 of 8375 Old 11-19-2019, 05:01 PM
AVS Forum Special Member
 
Manni01's Avatar
 
Join Date: Sep 2008
Posts: 9,410
Mentioned: 366 Post(s)
Tagged: 0 Thread(s)
Quoted: 5690 Post(s)
Liked: 6026
Quote:
Originally Posted by Soulnight View Post
@Manni01

I don't have an issue at all that you like the new one better.
I am not trying to convince you. Just reporting what I saw.

I would prefer to see the same thing that you do and that we agree but it's also ok when it does not happen.

I personally did not see any advantage of the new log from Neo-xp over the old log. Like I did not see any advantage of L-80 over L-63 before that.
Maybe you can provide some examples where new log has better contrast than old log.

For me, it's more like 99% of the time L-80 has a much better contrast than any Log while still feeling natural. In very rare cases, log may looks better.

In any case, I am happy that we have (still) now some good curves for different taste/ setup.

1) old log / new log (the safe setting, no matter what).
2) L-80 ( strong contrast)
3) Mirror (maximum pop)

And I do like "log" by the way. Old and new. It's a safe setting and still much better than HSTM off. I am just discussing preferences at this point. And it's important for me to also have the log in the final HSTM settings so that if my taste/ setup evolves, I can roll to the pure log instead.

I'm not trying to convince you either, only to reply to your statement "I don't understand all the excitement with the new curves".

The excitement comes from the fact that we're not seeing what you're seeing, or we have different taste, or a combination of both.

I suggested a simple way for you to possibly see where the excitement comes from, but it's fine if you're not interested. I'm not really interested in redoing all these tests with 50nits and settings that don't matter to me either

Choices are good, and I don't think that madshi has a problem with that.

I have already given detailed titles, timecodes and frame numbers where L80 here does very poorly compared to the latest curves, they are the same where new log does better than old log here, along with a few others (for example the sub pilot in The Meg). They are there for you to check. The issues might be different, or not present in your setup, so I wouldn't waste time looking for them unless you use my settings and not yours.

There is nothing wrong with preferring L-80/L-63. You might have missed it but I did say in one of my reports that I agreed with you and Anna and that there was little difference between the two curves.
Soulnight likes this.

Batch Utility V4.02 May 16 2019 to automate measurements files for madVR with support for BD Folders
JVC Macro feature on Vertex/Vertex2/Integral2/Maestro

Last edited by Manni01; 11-19-2019 at 06:17 PM.
Manni01 is offline  
post #8081 of 8375 Old 11-19-2019, 05:45 PM
Senior Member
 
Join Date: Feb 2019
Location: Iowa
Posts: 201
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 203 Post(s)
Liked: 133
Tested Neo-XP (2.0) @ 100 and @ 125; 3D effect not as deep and less contrast than Neo-XP (1.0) @ 125 or Fer15 (m38) @ 35. If I could only have one curve at this point, it would be Neo-XP (1.0) @ 125. But would prefer option of Log-80, Neo-XP (1.0), and Fer 15 (m38). Tested on LG 400 nit HDR (hdr format) and JVC RS540 (SDR2020).

The more I study Neo-XP (1.0) the more I appreciate the fluidity, smoothness, pop, and deep field of this curve.

But I always welcome more choices, than less.
Manni01 likes this.

Last edited by catav; 11-19-2019 at 05:54 PM.
catav is offline  
post #8082 of 8375 Old 11-19-2019, 06:39 PM
Advanced Member
 
jbn008's Avatar
 
Join Date: Sep 2007
Location: Charlotte, NC
Posts: 833
Mentioned: 5 Post(s)
Tagged: 0 Thread(s)
Quoted: 272 Post(s)
Liked: 144
I have a quick question and I apologize if it's been answered already. If we engage HSTM should we go back to a standard HDR gamma curve on our display? Before HSTM, my display (JVC) was set to bt.2020 with a SDR 2.4 gamma to match my madvr settings. Thanks
jbn008 is online now  
post #8083 of 8375 Old 11-19-2019, 08:09 PM
AVS Forum Addicted Member
 
lovingdvd's Avatar
 
Join Date: Jul 2002
Posts: 13,014
Mentioned: 91 Post(s)
Tagged: 0 Thread(s)
Quoted: 3742 Post(s)
Liked: 905
I spent some more time tonight trying to figure out what is going on with the clipping and hotspotting with HSTM in The Greatest Showman. DI is off. I tried with Neo-XP's two very latest curves at 100 strength and Manni's L80 at 50. All of them exhibit unwanted clipping / loss of detail and hotspotting in his jacket with HSTM on, and no such issues with HSTM off. So this is not related to new curves, but has been there the whole time, at least in my setup, in specific scenes.

So here's the thing - if I turn off the 3D LUT calibration then the issue goes away completely and there's no loss of detail and or hotspotting with HSTM on or off. So I wonder then what can be causing it. With calibration on the jacket is definitely a deeper/brighter red, but I believe this is to be expected. This is calibrated using CalMAN and similar results with DisplayCal, in terms of color.

@madshi - is it possible that the 3D LUT may need to take into account changes being made to the gamma? Just speculating. I'm not sure what the cause or proper fix is. I do believe the calibration to be accurate.

Attached is a picture I took very close to the screen which highlights the issues. You can see in the ON version the orange hotspots which are not in the OFF version, as well as an example of details in the texture of the jacket in the OFF version which disappear due to clipping in the ON version.

Very curious what you guys make of this. And for others that report clipping issues like @stefanelli73 can you please try turning the 3D LUT calibration off temporarily and see if that makes a difference? Thanks!
Attached Thumbnails
Click image for larger version

Name:	hcomp.png
Views:	51
Size:	1.57 MB
ID:	2643260  
lovingdvd is offline  
post #8084 of 8375 Old 11-19-2019, 08:43 PM
AVS Forum Special Member
 
SirMaster's Avatar
 
Join Date: Aug 2009
Location: Wisconsin
Posts: 1,004
Mentioned: 42 Post(s)
Tagged: 0 Thread(s)
Quoted: 845 Post(s)
Liked: 529
Quote:
Originally Posted by lovingdvd View Post
I spent some more time tonight trying to figure out what is going on with the clipping and hotspotting with HSTM in The Greatest Showman. DI is off. I tried with Neo-XP's two very latest curves at 100 strength and Manni's L80 at 50. All of them exhibit unwanted clipping / loss of detail and hotspotting in his jacket with HSTM on, and no such issues with HSTM off. So this is not related to new curves, but has been there the whole time, at least in my setup, in specific scenes.



So here's the thing - if I turn off the 3D LUT calibration then the issue goes away completely and there's no loss of detail and or hotspotting with HSTM on or off. So I wonder then what can be causing it. With calibration on the jacket is definitely a deeper/brighter red, but I believe this is to be expected. This is calibrated using CalMAN and similar results with DisplayCal, in terms of color.


@madshi - is it possible that the 3D LUT may need to take into account changes being made to the gamma? Just speculating. I'm not sure what the cause or proper fix is. I do believe the calibration to be accurate.



Attached is a picture I took very close to the screen which highlights the issues. You can see in the ON version the orange hotspots which are not in the OFF version, as well as an example of details in the texture of the jacket in the OFF version which disappear due to clipping in the ON version.



Very curious what you guys make of this. And for others that report clipping issues like @stefanelli73 can you please try turning the 3D LUT calibration off temporarily and see if that makes a difference? Thanks!


What is your projector again?

If you make a 3DLUT for a color gamut that’s bigger than your projectors native color gamut, you can get posterization which looks like clipping.

Using more patches for the 3DLUT seems to help reduce this at least with DisplayCAL.

You could try looking at color clipping test patterns like a 0-255 gradient of red and see if you see a difference with the 3DLUT on compared to off.

HSTM might just be pushing that scene on your projector closer to the red gamut edge where you are clipping.
SirMaster is online now  
post #8085 of 8375 Old 11-19-2019, 09:58 PM
AVS Forum Addicted Member
 
lovingdvd's Avatar
 
Join Date: Jul 2002
Posts: 13,014
Mentioned: 91 Post(s)
Tagged: 0 Thread(s)
Quoted: 3742 Post(s)
Liked: 905
Quote:
Originally Posted by SirMaster View Post
What is your projector again?

If you make a 3DLUT for a color gamut that’s bigger than your projectors native color gamut, you can get posterization which looks like clipping.

Using more patches for the 3DLUT seems to help reduce this at least with DisplayCAL.

You could try looking at color clipping test patterns like a 0-255 gradient of red and see if you see a difference with the 3DLUT on compared to off.

HSTM might just be pushing that scene on your projector closer to the red gamut edge where you are clipping.
Thanks! That sounds entirely plausable. I have the JVC RS500. I'm using @Manni01 's DCI P3 color profile preset without the filter, calibrated to DCI P3 with D65 color space. I used around 1200 patches for the 3D LUT calibration in both CalMAN and DisplayCal. I wonder if its a case where its trying to push red toward the edge of the gamut in those areas, but because there's no color filter it simply runs out of room, and hence the clipping? I wonder if a PCI P3 calibration WITH the filter would make a difference here? I'm starved for nits at just 45, so running with the filter isn't an option, but I am curious if that would solve it. Thoughts on that?

On a related but separate note - a month or so ago after a 3D LUT calibration with DisplayCal I noticed banding in staging lighting in certain scenes in chapter 7 of A Star Is Born. IIRC it was in the blues not reds. I eventually determined that turning off the 3D LUT resolved the banding. I then redid the calibration with CalMAN and did not get the banding that way. I reported the issue to the DisplayCal developer, who thought it may be an issue with the JVC internal processing and something related to Argyll. Anyway because the issue did not happen with CalMAN and I didn't have any more time to spend on it I just put that aside and stuck with the CalMAN calibration, and not sure if this is solvable with DisplayCal and how if so. Anyway I bring it up because it seems somewhat related. As mentioned I had done 1200 or so patches in DisplayCal (and CalMAN) which was around 2.5 hours of calibration time. I wonder if even a higher count like 2500 would make a difference here?
lovingdvd is offline  
post #8086 of 8375 Old 11-19-2019, 10:11 PM
AVS Forum Special Member
 
SirMaster's Avatar
 
Join Date: Aug 2009
Location: Wisconsin
Posts: 1,004
Mentioned: 42 Post(s)
Tagged: 0 Thread(s)
Quoted: 845 Post(s)
Liked: 529
Quote:
Originally Posted by lovingdvd View Post
Thanks! That sounds entirely plausable. I have the JVC RS500. I'm using @Manni01 's DCI P3 color profile preset without the filter, calibrated to DCI P3 with D65 color space. I used around 1200 patches for the 3D LUT calibration in both CalMAN and DisplayCal. I wonder if its a case where its trying to push red toward the edge of the gamut in those areas, but because there's no color filter it simply runs out of room, and hence the clipping? I wonder if a PCI P3 calibration WITH the filter would make a difference here? I'm starved for nits at just 45, so running with the filter isn't an option, but I am curious if that would solve it. Thoughts on that?

On a related but separate note - a month or so ago after a 3D LUT calibration with DisplayCal I noticed banding in staging lighting in certain scenes in chapter 7 of A Star Is Born. IIRC it was in the blues not reds. I eventually determined that turning off the 3D LUT resolved the banding. I then redid the calibration with CalMAN and did not get the banding that way. I reported the issue to the DisplayCal developer, who thought it may be an issue with the JVC internal processing and something related to Argyll. Anyway because the issue did not happen with CalMAN and I didn't have any more time to spend on it I just put that aside and stuck with the CalMAN calibration, and not sure if this is solvable with DisplayCal and how if so. Anyway I bring it up because it seems somewhat related. As mentioned I had done 1200 or so patches in DisplayCal (and CalMAN) which was around 2.5 hours of calibration time. I wonder if even a higher count like 2500 would make a difference here?
Hmm, I think the filter would fix the issue, but it's not worth it with your nits.

You could consider creating a rec709 3DLUT and outputting madVR as rec709. Yes you lose a little color gamut, but I think it's probably worth it to retain your brightness. You don't really reach P3 anyways without the filter so you aren't losing all that much. You might be surprised how good HDR movies still look converted to rec709.

I ran into this banding and posterization effect when making 3DLUTs for my NX5 especially when trying to make one for BT2020. I did notice that using more patches lessened the effect, but I did not need very many patches. I was doing quick 3DLUTs like 115 or 175 patches and when I did an 485 and 895 patch 3DLUT, the posterization was pretty much gone.

After that I used P3 gamut profile rather than BT2020 and the posterization was even less with even 175 patches being enough to produce a result in which I did not notice any obvious posterization.
SirMaster is online now  
post #8087 of 8375 Old 11-19-2019, 11:37 PM
Member
 
RioBar4U's Avatar
 
Join Date: Dec 2011
Location: Switzerland
Posts: 88
Mentioned: 6 Post(s)
Tagged: 0 Thread(s)
Quoted: 25 Post(s)
Liked: 39
Red face

Also Mori and me did a next test session last night.

I was from the begining allways on the conservativ side and never liked the stronger curves like L80 or mirror, which I don't like at all. For me NeoXp last curve from sunday (Neo-XP
(strength 100): 1000-1050-1100-1150-1250-1350-1500-1750-2000-2000-2000-2000-2000-2000-2000-2000) was the best compromise so far. It's pushy, but never goes to far and produces the less 'errors' in the picture.
The new curves from Fer15 and NeoXP look the same for me. They are more aggressiv as Neo XP's from sunday and I can understand that peolpe like it, but for my taste it's too much.

So I would wish to keep the Neo's Neo-XP
(strength 100): 1000-1050-1100-1150-1250-1350-1500-1750-2000-2000-2000-2000-2000-2000-2000-2000 as a low strengh curve.

My suggestion:

Low HSTM: Neo-XP (strength 100): 1000-1050-1100-1150-1250-1350-1500-1750-2000-2000-2000-2000-2000-2000-2000-2000
Mid HSTM: new Fer15/Neo XP
Srong HSTM: L80 or Mirror or an other aggressiv curve, which people wish.
*Mori* and catav like this.
RioBar4U is online now  
post #8088 of 8375 Old 11-20-2019, 12:01 AM
Member
 
stefanelli73's Avatar
 
Join Date: Jan 2008
Posts: 115
Mentioned: 6 Post(s)
Tagged: 0 Thread(s)
Quoted: 32 Post(s)
Liked: 19
Quote:
Originally Posted by lovingdvd View Post
I spent some more time tonight trying to figure out what is going on with the clipping and hotspotting with HSTM in The Greatest Showman. DI is off. I tried with Neo-XP's two very latest curves at 100 strength and Manni's L80 at 50. All of them exhibit unwanted clipping / loss of detail and hotspotting in his jacket with HSTM on, and no such issues with HSTM off. So this is not related to new curves, but has been there the whole time, at least in my setup, in specific scenes.

So here's the thing - if I turn off the 3D LUT calibration then the issue goes away completely and there's no loss of detail and or hotspotting with HSTM on or off. So I wonder then what can be causing it. With calibration on the jacket is definitely a deeper/brighter red, but I believe this is to be expected. This is calibrated using CalMAN and similar results with DisplayCal, in terms of color.

@madshi - is it possible that the 3D LUT may need to take into account changes being made to the gamma? Just speculating. I'm not sure what the cause or proper fix is. I do believe the calibration to be accurate.

Attached is a picture I took very close to the screen which highlights the issues. You can see in the ON version the orange hotspots which are not in the OFF version, as well as an example of details in the texture of the jacket in the OFF version which disappear due to clipping in the ON version.

Very curious what you guys make of this. And for others that report clipping issues like @stefanelli73 can you please try turning the 3D LUT calibration off temporarily and see if that makes a difference? Thanks!

yesterday I tried the two curves, both that of NEO-XP and FER15, both are very similar, so it is difficult to choose, at the moment I'm using the FER15 one, as it returns, a more "pop" image ... ... concerning the red clipping, repeatedly reported by me, on the films I mentioned a few posts above, there is still .... the scenes to which I refer in ATOMIC BLONDE, are those of the bar and the disco, if can you find them or do I provide the exact minutes today .... as I have always said by deactivating HTSM the clipping problem disappears, but actually I have not tried to disable 3dlut in DCI-P3, nor have I made one in SDR-BT.2020 and in BT.709, I have to see if it also happens with these
stefanelli73 is offline  
post #8089 of 8375 Old 11-20-2019, 03:26 AM
AVS Forum Special Member
 
Manni01's Avatar
 
Join Date: Sep 2008
Posts: 9,410
Mentioned: 366 Post(s)
Tagged: 0 Thread(s)
Quoted: 5690 Post(s)
Liked: 6026
Quote:
Originally Posted by lovingdvd View Post
I spent some more time tonight trying to figure out what is going on with the clipping and hotspotting with HSTM in The Greatest Showman. DI is off. I tried with Neo-XP's two very latest curves at 100 strength and Manni's L80 at 50. All of them exhibit unwanted clipping / loss of detail and hotspotting in his jacket with HSTM on, and no such issues with HSTM off. So this is not related to new curves, but has been there the whole time, at least in my setup, in specific scenes.

So here's the thing - if I turn off the 3D LUT calibration then the issue goes away completely and there's no loss of detail and or hotspotting with HSTM on or off. So I wonder then what can be causing it. With calibration on the jacket is definitely a deeper/brighter red, but I believe this is to be expected. This is calibrated using CalMAN and similar results with DisplayCal, in terms of color.

@madshi - is it possible that the 3D LUT may need to take into account changes being made to the gamma? Just speculating. I'm not sure what the cause or proper fix is. I do believe the calibration to be accurate.

Attached is a picture I took very close to the screen which highlights the issues. You can see in the ON version the orange hotspots which are not in the OFF version, as well as an example of details in the texture of the jacket in the OFF version which disappear due to clipping in the ON version.

Very curious what you guys make of this. And for others that report clipping issues like @stefanelli73 can you please try turning the 3D LUT calibration off temporarily and see if that makes a difference? Thanks!
As discussed already, using a small LUT is only possible if you 1) cover 100% of the target gamut and 2) use the JVC Autocal to get a good baseline (under 5 dE max, 2-3 dE average). Without the filter, you can't use a 3D LUT unless you do at least a 5,000 points LUT, simply because you are way too far from the targets as you only cover 90% of P3 at best.

Your Calman LUT might not have shown the same issue on blue, but it shows one on red.

A bigger LUT might alleviate the problem, but it will still likely be there.

I agree that using a rec-709 3D LUT (if you cover 100% of rec-709 or close without the filter) would probably be a better solution than using the filter and drop 10-20% brightness.

You're not getting much more than rec-709 anyway without the filter, so would get better results that way likely.

Batch Utility V4.02 May 16 2019 to automate measurements files for madVR with support for BD Folders
JVC Macro feature on Vertex/Vertex2/Integral2/Maestro
Manni01 is offline  
post #8090 of 8375 Old 11-20-2019, 03:53 AM
Senior Member
 
*Mori*'s Avatar
 
Join Date: Feb 2009
Posts: 302
Mentioned: 24 Post(s)
Tagged: 0 Thread(s)
Quoted: 319 Post(s)
Liked: 485
As RioBar4U wrote we compared HSTM Off (1, control) and Mirror (5 ) to

Quote:
(3) Neo-XP (the one that I liked a lot)
1000-1050-1100-1150-1250-1350-1500-1750-2000-2000-2000-2000-2000-2000-2000-2000

(4) Neo-XP Pure Log, strength 100: (relatively new)
750-1019-1134-1202-1287-1343-1403-1471-1519-1556-1586-1612-1634-1672-1702-1750

(5) Fer15 (Neu), strength 35 : (relatively new)
380-880-1280-1590-2070-2400-2760-3140-3380-3540-3660-3750-3820-3930-4000-4100
The new versions by Neo-Xp and Fer15 (4,5) were very much the same to me. Hardly any difference and I could'nt say which I'd prefer. They showed sometimes a little bit more 3D effect (but not always) than my preferred Neo-XP - sometimes less. The effect of HSTM became certainly more obvious here compared to (3). Mirrow and HSTM off were always used as positive (Mirror) or negative controls (off). Like I said before also in my view there is room for more 3D between that Neo-XP curve and Mirror. Both new curves (4,5) fullfill that job IMO.

My favorite is still (3). Sometimes brighter elements were at the border to be too strong. But never too much. And I got the HSTM effect still very nicely compared to off (1). Shadow details were often better than (4,5). For me I would use that as my standard curve. Very balanced. I wouldn't use Mirror or L80 (which didn't work at all for me). The clear winner for me was (3) again. It is never off compared to others and still providing the benefits compared to HSTM off. All that reminds me of the Sony Contrast Enhancer discussions. I may be considered as a purist since I hardly used it. But I hope not be seen as a fundamentalist because I see why and in which setup people use the Contrast Enhancer for a good reason. L80 and Mirrow in my view are considerbly worse than the 2 Sony Contrast enhancer modes, but (4,5) are better IMO. And I have been using Sony projectors for many years

Nevertheless Mirror is a curve I would like to see in Envy mainly because there are hardly any alternatives on the more aggressive side. I would like to say here that both Neo-XP and Fer15 did a great job for improving the curves which were a bit more conservative but still showing HSTM effect without beeing obviously far too much (for some people like me). I think, from a neutral perspective, it would be still worthwhile to find something below Mirror. Likely it will still too much for me but my impression is that lots of work has been invested on the lower (conservative) side but quite little has been done on the higher (progressive) side.

P.S. another friend of mine is setting up a PC incl. madVR these days. Since he has the same screen and projector like I have I will have then the oportunity to test on a considerably brighter setup (180 vs 80 Nits) as well.
Manni01, neo_2009 and RioBar4U like this.

Last edited by *Mori*; 11-20-2019 at 04:03 AM.
*Mori* is offline  
post #8091 of 8375 Old 11-20-2019, 05:02 AM
Member
 
stefanelli73's Avatar
 
Join Date: Jan 2008
Posts: 115
Mentioned: 6 Post(s)
Tagged: 0 Thread(s)
Quoted: 32 Post(s)
Liked: 19
@lovingdvd

The scenes of ATOMIC BLONDE concerning the clipping of the red are from the minute 36 ': 57 "to 40': 16" (bar scene) and from 50 ': 57 to 52': 40 "(disco scene) .... then the problem disappears only if I go to THIS DISPLAY IS ALREADY CALIBRATED, choose DCI-P3 and pure power curve 2.20, instead with any 3dlut (BT.709, SDR DCI-P3 and SDR BT.2020) the problem remains.

Last edited by stefanelli73; 11-20-2019 at 05:29 AM.
stefanelli73 is offline  
post #8092 of 8375 Old 11-20-2019, 05:18 AM
Newbie
 
Join Date: Oct 2019
Posts: 10
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 3 Post(s)
Liked: 1
For me personally, I don't need the absolutely perfect, natural picture quality. I want to have fun when I'm watching a movie. And fun means to me in this case, 3D effect. If there is, like the L-80 curve, sometimes a little bit of unnatural colour, that's okay for me, if I will have a very depth feeling picture. To be honest, I don't used the new curves so far, so I can't say if the 3d effect is not that good like in the L-80 curve. Can somebody tell me the actual 3d effect generated by the new curves? THX in advance!
SmoothR is offline  
post #8093 of 8375 Old 11-20-2019, 05:32 AM
Advanced Member
 
chros73's Avatar
 
Join Date: Jan 2015
Posts: 617
Mentioned: 12 Post(s)
Tagged: 0 Thread(s)
Quoted: 385 Post(s)
Liked: 178
Quote:
Originally Posted by Onkyoman View Post
When HDR output is set, is HSTM tone mapping touching the entire frame when it goes above 700 nits or just the compressed area of the display curve from around 400-700 nits?
It touches the whole frame.

Quote:
Originally Posted by Onkyoman View Post
If it is touching the entire frame, is this necessary when the display can do a linear representation of 0-100 nits with accurate PQ EOTF tracking 100% of the time?
I'm not sure we can talk about "0-100 nits" in the traditional (SDR) sense regarding to HDR. (e.g. The Meg as an example with its 1193 nits maxFALL.)
But interestingly, Vincent Teoh mentioned multiple times in his Panasonic OLED reviews that Panasonic's DTM doesn't brighten/darken the whole frame (only affects the highlights), unlike LG's DTM.
But personally I don't have any issue with touching the whole frame even with high real display peak nits either.

Quote:
Originally Posted by Onkyoman View Post
Is this setting really necessary? All HDR displays will be expecting HDR sources to have BT.2020 gamut coordinates because the PQ HDR10 standard requires that HDR10 sources are mastered with BT.2020 primaries.
We don't know, until we try/test it. Maybe it will, maybe it won't. The internal processing of these displays is really complicated, especially when it comes to HDR.
Do you have an OLED display yet?

Quote:
Originally Posted by Onkyoman View Post
you typically have to enable the DCI-P3 gamut mode manually
Yes, that's the idea, but it *can* (and not necessarily will) only work if madvr does tone mapping with "output HDR" on.

Ryzen 5 2600,Asus Prime b450-Plus,16GB,MSI GTX 1060 Gaming X 6GB(v385.28),Win10 LTSB 1607,MPC-BEx64+LAV+MadVR,Yamaha RX-A870,LG OLED65B8(04.10.25+PC4:4:[email protected]/24/25/29/30/50/59/60Hz)

Last edited by chros73; 11-20-2019 at 05:37 AM.
chros73 is offline  
post #8094 of 8375 Old 11-20-2019, 07:50 AM
AVS Forum Special Member
 
Onkyoman's Avatar
 
Join Date: Aug 2007
Location: Canada
Posts: 1,958
Mentioned: 54 Post(s)
Tagged: 0 Thread(s)
Quoted: 1148 Post(s)
Liked: 678
Quote:
Originally Posted by chros73 View Post
It touches the whole frame.

I'm not sure we can talk about "0-100 nits" in the traditional (SDR) sense regarding to HDR. (e.g. The Meg as an example with its 1193 nits maxFALL.)
But interestingly, Vincent Teoh mentioned multiple times in his Panasonic OLED reviews that Panasonic's DTM doesn't brighten/darken the whole frame (only affects the highlights), unlike LG's DTM.
But personally I don't have any issue with touching the whole frame even with high real display peak nits either.
The HDR display owners should determine whether they prefer whole scene tone mapping versus simply compressing the specular highlights back into display range. LG's dynamic tone mapping sounds too excessive in how it manipulates the image. But it is probably not that far off how madVR's HSTM works in the sense that it emphasizes presering the relative contrast of bright HDR scenes more so than preserving the APL or accurate PQ EOTF tracking. On projector, that is necessary because accurate PQ EOTF tracking up to 100 nits is rarely ever going to work without looking completely clipped or flat and you have to make more significant compromises to balance contrast and brightness.

I do question whether this enhancement is necessary when your display can output 700 nits if ABL isn't involved and the display has availalbe nits between 100 nits and 700 nits to roll-off any specular highlights in any movie. madVR manages to do the same trick with less than 50 nits in some cases and still manages to minimze visible highlight clipping of specular highlights mastered as high as 10,000 nits. It seems like a waste of processing to compress 90% of the picture when you have as much as 600 nits to work with on OLED TVs to deal with only the specular highlights.

I know The Meg is mastered extremely bright, but those high APLs are still considered "specular highlights" in PQ HDR terms. Not many movies are going to be mastered with that much content above 100 nits when those brightness levels are supposed to be reserved for highlight peaks. The Meg primarily takes place outdoors, so those high APLs can somewhat be justified. The majority of HDR10 movies measured with madMeasureHDR have less than 10% of the pixels above 100 nits and many movies only use 5% of the brightness levels for specular highlights. That would be considered a "normal" use of the PQ HDR luminance scale.

Most HDR displays that value accuracy, particularly Panasonic, Sony and LG (static tone mappping only), all follow the PQ EOTF 1:1 up to at least 100 nits before starting the roll-off point. Some LEDs from manufacturers such as Samsung and TCL manage to linearly brighten the entire PQ EOTF to make HDR content look brighter on LEDs than it does on OLED displays. This seems more dishonest and solely a marketing ploy, not a proper way to represent PQ HDR content.

Is it better to preserve accurate PQ EOTF tracking with accurate PQ luminance output for most source values (0-100 nits) for all scenes and only compress the specular highlights in brighter scenes, or it is better to maninpulate 0-100 nits for scenes above 700 nits by giving the image extra contrast by manipulating the midtones and shadows? The first approach would be more accurate and second approach might look better in a few isolated scenes. Or it could even look distracting or worse by suddenly shifting the APL of the overall image in the middle of a scene.

Quote:
Originally Posted by chros73 View Post
We don't know, until we try/test it. Maybe it will, maybe it won't. The internal processing of these displays is really complicated, especially when it comes to HDR.
Do you have an OLED display yet?

Yes, that's the idea, but it *can* (and not necessarily will) only work if madvr does tone mapping with "output HDR" on.
I don't have an OLED display to test. Maybe it would work to send DCI-P3 primaries to your TV. The display might automatically switch to an internal DCI-P3 color profile. But you are trying to watch HDR movies with a display mode that was originally designed on most displays for color grading, not watching HDR movies.

As I said, BT.2020 is the correct gamut for HDR10 content. The option to use BT.2020 or DCI-P3 color processing with SDR display modes is there for those that want to upconvert legacy SDR content to take advantage of the wider color gamut of most UHD displays. HDR content is already using the widest gamut of the display. The display video processing is to blame if it is creating errors when processing a native HDR source with correct BT.2020 source primaries and accurate HDR10 metadata.
sat4all likes this.
Onkyoman is offline  
post #8095 of 8375 Old 11-20-2019, 09:45 AM
AVS Forum Addicted Member
 
lovingdvd's Avatar
 
Join Date: Jul 2002
Posts: 13,014
Mentioned: 91 Post(s)
Tagged: 0 Thread(s)
Quoted: 3742 Post(s)
Liked: 905
Quote:
Originally Posted by SirMaster View Post
Hmm, I think the filter would fix the issue, but it's not worth it with your nits.
Agreed.

Quote:
You could consider creating a rec709 3DLUT and outputting madVR as rec709. Yes you lose a little color gamut, but I think it's probably worth it to retain your brightness. You don't really reach P3 anyways without the filter so you aren't losing all that much. You might be surprised how good HDR movies still look converted to rec709.
If there is even a small but noticeable benefit to using P3 without the filter compared to using BT709, I think I still would prefer to stick with P3 for that marginal improvement (again, assuming it is noticeable, despite being marginal). That's because AFAIU ATM the clipping issue I'm seeing is on just a few rare and specific scenes. For instance just to have his red coat like perfect in Great Showman and to not have some clipping in Atomic Blonde (I haven't checked for clipping here yet) that's not enough motivation to move to BT709. Now as I pay more attention to the issue, if I start noticing it in more and more places, with a fair number of movies, and it's not as isolated as I think it is currently, then I may try BT709.

Quote:
I ran into this banding and posterization effect when making 3DLUTs for my NX5 especially when trying to make one for BT2020. I did notice that using more patches lessened the effect, but I did not need very many patches. I was doing quick 3DLUTs like 115 or 175 patches and when I did an 485 and 895 patch 3DLUT, the posterization was pretty much gone.

After that I used P3 gamut profile rather than BT2020 and the posterization was even less with even 175 patches being enough to produce a result in which I did not notice any obvious posterization.
OK that may be a difference between my RS500 and your NX5 which is several (3 I think) versions newer.

Quote:
Originally Posted by Manni01 View Post
As discussed already, using a small LUT is only possible if you 1) cover 100% of the target gamut and 2) use the JVC Autocal to get a good baseline (under 5 dE max, 2-3 dE average). Without the filter, you can't use a 3D LUT unless you do at least a 5,000 points LUT, simply because you are way too far from the targets as you only cover 90% of P3 at best.
Oh wow, I didn't realize we were talking about THAT much larger. I thought 1200 points was already pretty big. I think 5,000 points would take CalMAN and DisplayCal maybe like 6-8 hours to complete. Could be worth a try, maybe just let it run overnight. If anything just to see out of curosity if it improves things.

Post calibration the CalMAN color checker showed exceptional results with the 1200 patches for all measures, except the ones at the edge of the gamut, which is what would be expected. But of course that doesn't sound like "good enough" for handling these particular "edge" cases in movies.

Quote:
Your Calman LUT might not have shown the same issue on blue, but it shows one on red.

A bigger LUT might alleviate the problem, but it will still likely be there.

I agree that using a rec-709 3D LUT (if you cover 100% of rec-709 or close without the filter) would probably be a better solution than using the filter and drop 10-20% brightness.

You're not getting much more than rec-709 anyway without the filter, so would get better results that way likely.
As mentioned above, it's hard to know whether its worth moving to BT709 or not without knowing how much of a noticeable difference there is with BT709 vs DCI P3 without the filter AND without knowing to what extent I am having these edge cases like with the jacket in Greatest Showman. If BT709 vs P3 with no filter is virtually indistinguishable then sure, why not. But I assume there's still some benefit; otherwise it seems many people who use DCI P3 without the filter would just instead use BT709 in the first place? Unless those people are not using madVR so they don't have a choice to have excellent color space mapping...? And likewise as also mentioned, if these are just a few isolated cases, and there still is SOME benefit to DCI P3 without the filter, I'd probably want to just stick with that.

Also just to make sure - AFAIU, if I wanted to try BT709 I would just do a fresh calibration with the PJ set for BT709 and then remove the DCI P3 calibration from madVR, so the only calibration file it has is the newly created BT709 one. Then when I tell madVR to use the 3D LUT calibration it will see there is only the BT709 one, and will automatically map BT2020 to BT709 and apply that 3D LUT, yes?

Then again - I'm not sure it's even worth the trouble. See below where @stefanelli73 says he has the issue with any 3D LUT including BT709.

Quote:
Originally Posted by stefanelli73 View Post
@lovingdvd

The scenes of ATOMIC BLONDE concerning the clipping of the red are from the minute 36 ': 57 "to 40': 16" (bar scene) and from 50 ': 57 to 52': 40 "(disco scene) .... then the problem disappears only if I go to THIS DISPLAY IS ALREADY CALIBRATED, choose DCI-P3 and pure power curve 2.20, instead with any 3dlut (BT.709, SDR DCI-P3 and SDR BT.2020) the problem remains.
Thanks. Unfortunately this seems to confirm my findings as well, except I haven't tried with BT709 and probably won't go through that trouble considering you say it doesn't help. Just to be sure I understand - you are saying that even when tone mapping to BT709 and using a targetted BT709 calibration (and not a re-targeted on from a DCI P3 or BT2020 calibration) you have the same problem with the clipping and over-saturation?
lovingdvd is offline  
post #8096 of 8375 Old 11-20-2019, 10:05 AM
AVS Forum Special Member
 
Manni01's Avatar
 
Join Date: Sep 2008
Posts: 9,410
Mentioned: 366 Post(s)
Tagged: 0 Thread(s)
Quoted: 5690 Post(s)
Liked: 6026
Quote:
Originally Posted by lovingdvd View Post
Agreed.



If there is even a small but noticeable benefit to using P3 without the filter compared to using BT709, I think I still would prefer to stick with P3 for that marginal improvement (again, assuming it is noticeable, despite being marginal). That's because AFAIU ATM the clipping issue I'm seeing is on just a few rare and specific scenes. For instance just to have his red coat like perfect in Great Showman and to not have some clipping in Atomic Blonde (I haven't checked for clipping here yet) that's not enough motivation to move to BT709. Now as I pay more attention to the issue, if I start noticing it in more and more places, with a fair number of movies, and it's not as isolated as I think it is currently, then I may try BT709.



OK that may be a difference between my RS500 and your NX5 which is several (3 I think) versions newer.



Oh wow, I didn't realize we were talking about THAT much larger. I thought 1200 points was already pretty big. I think 5,000 points would take CalMAN and DisplayCal maybe like 6-8 hours to complete. Could be worth a try, maybe just let it run overnight. If anything just to see out of curosity if it improves things.

Post calibration the CalMAN color checker showed exceptional results with the 1200 patches for all measures, except the ones at the edge of the gamut, which is what would be expected. But of course that doesn't sound like "good enough" for handling these particular "edge" cases in movies.



As mentioned above, it's hard to know whether its worth moving to BT709 or not without knowing how much of a noticeable difference there is with BT709 vs DCI P3 without the filter AND without knowing to what extent I am having these edge cases like with the jacket in Greatest Showman. If BT709 vs P3 with no filter is virtually indistinguishable then sure, why not. But I assume there's still some benefit; otherwise it seems many people who use DCI P3 without the filter would just instead use BT709 in the first place? Unless those people are not using madVR so they don't have a choice to have excellent color space mapping...? And likewise as also mentioned, if these are just a few isolated cases, and there still is SOME benefit to DCI P3 without the filter, I'd probably want to just stick with that.

Also just to make sure - AFAIU, if I wanted to try BT709 I would just do a fresh calibration with the PJ set for BT709 and then remove the DCI P3 calibration from madVR, so the only calibration file it has is the newly created BT709 one. Then when I tell madVR to use the 3D LUT calibration it will see there is only the BT709 one, and will automatically map BT2020 to BT709 and apply that 3D LUT, yes?

Then again - I'm not sure it's even worth the trouble. See below where @stefanelli73 says he has the issue with any 3D LUT including BT709.



Thanks. Unfortunately this seems to confirm my findings as well, except I haven't tried with BT709 and probably won't go through that trouble considering you say it doesn't help. Just to be sure I understand - you are saying that even when tone mapping to BT709 and using a targetted BT709 calibration (and not a re-targeted on from a DCI P3 or BT2020 calibration) you have the same problem with the clipping and over-saturation?
I don't really want to get into calibration issues here because it would take the thread off track.

I would suggest two simple tests:

1) In the calibration tab, select "this display is already calibrated" and specify rec-709 and the gamma you are using for your rec-709 baseline (before 3D LUT), and select the rec-709 baseline in the PJ (assuming it's been autocaled and is fairly accurate). Do you notice and color missed compared to selecting DCI-P3 in that same tab (with the gamma in your P3 baseline) and selecting your baseline P3 calibration, also I assume fairly accurate after autocal? As you're not using the filter in your P3 baseline, I assume the difference will be minimal.

2) Once you've established whether P3 without filter gives you anything from a color point of view vs rec-709, it's easier to decide if it makes sense or not to do a 3D LUT for rec-709 and use that as a target for your tonemapping (switching back to using a 3D LUT and having only the rec-709 LUT populated in the calibration tab).

If you can reach 100% of rec-709 without the filter, you should be able to get near reference results after the autocal with a Lightning LUT (around 100 points) that takes only 10 minutes or so. This is what I do for both rec-709 and P3 (as I use the filter and reach 97% of the gamut). Much better than wasting hours overnight to generate large LUTs, especially given how quickly these PJ drift and how often you need to update the LUTs (about every 200 hours here).

Good luck!
the_jaguar likes this.

Batch Utility V4.02 May 16 2019 to automate measurements files for madVR with support for BD Folders
JVC Macro feature on Vertex/Vertex2/Integral2/Maestro

Last edited by Manni01; 11-20-2019 at 10:21 AM.
Manni01 is offline  
post #8097 of 8375 Old 11-20-2019, 10:10 AM
Member
 
stefanelli73's Avatar
 
Join Date: Jan 2008
Posts: 115
Mentioned: 6 Post(s)
Tagged: 0 Thread(s)
Quoted: 32 Post(s)
Liked: 19
Quote:
Originally Posted by lovingdvd View Post

Thanks. Unfortunately this seems to confirm my findings as well, except I haven't tried with BT709 and probably won't go through that trouble considering you say it doesn't help. Just to be sure I understand - you are saying that even when tone mapping to BT709 and using a targetted BT709 calibration (and not a re-targeted on from a DCI P3 or BT2020 calibration) you have the same problem with the clipping and over-saturation?

I have realized nr. 3, 3dlut (1536 points for each of them) one in BT.709 for films 2k and two in SDR-DCI-P3 and SDR BT.2020 for 4k films .... from tests performed today if I use one of these 3dlut I have clipping ONLY RED COLOR, for some films (ATOMIC BLONDE, X-MEN APOCALYPSE, THE GREATEST SHOWMAN and GHOSTBUSTERS (2016)) not all, while if I use THIS DISPLAY IS ALREADY CALIBRATED and choose DCI-P3 and PURE POWER CURVE 2.20, the problem of red clipping disappears, same thing if I use BT.2020, but too much desaturate, while in BT.709 the problem is there.

Last edited by stefanelli73; 11-20-2019 at 10:36 AM.
stefanelli73 is offline  
post #8098 of 8375 Old 11-20-2019, 10:59 AM
AVS Forum Addicted Member
 
lovingdvd's Avatar
 
Join Date: Jul 2002
Posts: 13,014
Mentioned: 91 Post(s)
Tagged: 0 Thread(s)
Quoted: 3742 Post(s)
Liked: 905
Quote:
Originally Posted by Manni01 View Post
I don't really want to get into calibration issues here because it would take the thread off track.

I would suggest two simple tests:

1) In the calibration tab, select "this display is already calibrated" and specify rec-709 and the gamma you are using for your rec-709 baseline (before 3D LUT), and select the rec-709 baseline in the PJ (assuming it's been autocaled and is fairly accurate). Do you notice and color missed compared to selecting DCI-P3 in that same tab (with the gamma in your P3 baseline) and selecting your baseline P3 calibration, also I assume fairly accurate after autocal? As you're not using the filter in your P3 baseline, I assume the difference will be minimal.

2) Once you've established whether P3 without filter gives you anything from a color point of view vs rec-709, it's easier to decide if it makes sense or not to do a 3D LUT for rec-709 and use that as a target for your tonemapping (switching back to using a 3D LUT and having only the rec-709 LUT populated in the calibration tab).

If you can reach 100% of rec-709 without the filter, you should be able to get near reference results after the autocal with a Lightning LUT (around 100 points) that takes only 10 minutes or so. This is what I do for both rec-709 and P3 (as I use the filter and reach 97% of the gamut). Much better than wasting hours overnight to generate large LUTs, especially given how quickly these PJ drift and how often you need to update the LUTs (about every 200 hours here).

Good luck!
Thanks. I follow everything you're saying and will give it a try. There is a fair amount of (desirable) red push that I get with DCI P3 with the 3D LUT calibration compared to it off. I wonder if that's normal. In comparison it looks richer and more accurate with this push, which is why I say "desirable" because normally red push is considered a bad thing. Anyway back on track...

I think perhaps the key point not to lose sight of in all this is that the issue with the red clipping only occurs with HSTM on (seemingly any curve and at any strength). Therefore I wonder if it may be a good idea and appropriate for @madshi to add some sort of limiter option, that when enabled madVR would be extra careful not to push colors too bright when already near the edge of the gamut, perhaps even limited to just red. Maybe it could even have a % or some value so we can find the right threshold where it should not push past, which likely would be different for each user. Then again, Madshi may see this as a shortcoming of the projector and or the calibration, and that it's the responsiblity of the pj/calibration to be fully accurate instead of madVR acting as a bandaid in such cases.
lovingdvd is offline  
post #8099 of 8375 Old 11-20-2019, 11:01 AM
AVS Forum Addicted Member
 
lovingdvd's Avatar
 
Join Date: Jul 2002
Posts: 13,014
Mentioned: 91 Post(s)
Tagged: 0 Thread(s)
Quoted: 3742 Post(s)
Liked: 905
Quote:
Originally Posted by stefanelli73 View Post
I have realized nr. 3, 3dlut (1536 points for each of them) one in BT.709 for films 2k and two in SDR-DCI-P3 and SDR BT.2020 for 4k films .... from tests performed today if I use one of these 3dlut I have clipping ONLY RED COLOR, for some films (ATOMIC BLONDE, X-MEN APOCALYPSE, THE GREATEST SHOWMAN and GHOSTBUSTERS (2016)) not all, while if I use THIS DISPLAY IS ALREADY CALIBRATED and choose DCI-P3 and PURE POWER CURVE 2.20, the problem of red clipping disappears, same thing if I use BT.2020, but too much desaturate, while in BT.709 the problem is there.
Sorry but I didn't quite understand what you wrote above. Could you try rephrasing it a bit?

Also - are you saying that if you tell madVR to tone map to BT709 and use a calibrated BT709 setup that the Greatest Showman still shows the same red clipping / loss of details when madVR tone maps it to BT709 (as Manni suggests we try in his recent posts above)?
lovingdvd is offline  
post #8100 of 8375 Old 11-20-2019, 11:22 AM
Member
 
stefanelli73's Avatar
 
Join Date: Jan 2008
Posts: 115
Mentioned: 6 Post(s)
Tagged: 0 Thread(s)
Quoted: 32 Post(s)
Liked: 19
Quote:
Originally Posted by lovingdvd View Post
Sorry but I didn't quite understand what you wrote above. Could you try rephrasing it a bit?

Also - are you saying that if you tell madVR to tone map to BT709 and use a calibrated BT709 setup that the Greatest Showman still shows the same red clipping / loss of details when madVR tone maps it to BT709 (as Manni suggests we try in his recent posts above)?

I'm simply telling you that the RED clipping is present with any 3dlut I use (BT.709, DCI-P3 and BT.2020), if I don't use any 3dlut, but in Madvr's CALIBRATION I choose THIS DISPLAY IS ALREADY CALIBRATED and DCI-P3, the problem of clipping on red disappears.
stefanelli73 is offline  
Sponsored Links
Advertisement
 
Reply Digital Hi-End Projectors - $3,000+ USD MSRP

Tags
dynamic tone mapping , hdr , madvr , sdr , ton mapping

Thread Tools
Show Printable Version Show Printable Version
Email this Page Email this Page


Forum Jump: 

Posting Rules  
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off