AVS Forum banner

5501 - 5520 of 9458 Posts

·
Registered
Joined
·
1,209 Posts
Hey Neo, I have been working through the last 20ish pages trying to get an idea of how the settings have progressed with the builds. I was wondering on this post, "no compression limit" changed from 150 to 0 all of sudden (and still are using that value). Prior to this, you , Fer, and others were using 150. Why did you drop down to 0, was it based on something specific changing in the test builds?
That is just because I got "new" LCD monitor recently (for free) with lower nits peak than my other monitors/TV, and after testing it with 150 as "no compression limit", I preferred FALL algo all the way.

And since my OLED TV is calibrated to 150 nits, I use 0 for all madVR profiles now (easier).

Also is there a term definition list somewhere for the settings that are under HDR?
I don't think so, but if you have time, you can find all the answers in this thread: https://www.avsforum.com/forum/26-home-theater-computers/3040072-madvr-tool-madmeasurehdr-optimizer-measurements-dynamic-clipping-target-nits.html

@Neo-XP , new build 68, best settings please! :D
Best, but only for me and only for now:



The issues I still have are :

- Flashes causing big luminance adaptations (can't be fixed now, but hopefully in a future version with a "looking ahead" algo)
- Real scenes (but very few) not detected with the "subtraction of previous frame" option, because they happen just after frames with big metric1 and metric2 values.

I don't know what could be done for this one, even with the "looking ahead" algo (how to know it is a new scene after a few frames with big changes?).
 

·
Registered
Joined
·
1,479 Posts
Discussion Starter #5,502
Uuh, what's *that* you're talking about there!?!? :p

Edit: I'm really sorry, but I can't say much more about this yet. More info will come when I'm ready for that.

Edit2: Hopefully this thread will not be run over with speculation now, though?
Ok.... :eek:

Anyway, this was always my dream. The one thing always missing with madVR. :rolleyes:

Every time I talk about madVR, people are always saying: but can you play netflix or Apple TV or just TV through it... And always the response is no... And I had a look myself quite often with a capture video card but nothing seemed to be doing the job I needed.

I think I am just going to open a dedicated anticipation topic for this "dream" of mine.
MadVR with HDMI In.

Please post there if/when you can say more. :D
 

·
Registered
Joined
·
7,954 Posts
The issues I still have are :

- Flashes causing big luminance adaptations (can't be fixed now, but hopefully in a future version with a "looking ahead" algo)
- Real scenes (but very few) not detected with the "subtraction of previous frame" option, because they happen just after frames with big metric1 and metric2 values.

I don't know what could be done for this one, even with the "looking ahead" algo (how to know it is a new scene after a few frames with big changes?).
Yeah, the 2nd problem will be hard to crack. One possible option would be to try to improve the metrics so much that we don't have to substract the previous frame, anymore. But I'm not sure if that's reasonable. I also have some other ideas. But I'm taking this step by step, and the first step is to try make each metric on its own as good as possible.
 

·
Registered
Joined
·
1,479 Posts
Discussion Starter #5,506

·
Registered
Joined
·
146 Posts
I would like a suggestion on what settings to use on an OLED tv (suppose be a 700nits display). At least something to work with . Maybe some settings are not to use for an OLED\other HDR displays ? I guess dynamic clipping\Tonemap is also for hdr displays, but other then inserting 'real display nits' value and checking V dynamic clipping, is there any suggestion (even theoretically) on settings we might try?
 

·
Registered
Joined
·
466 Posts
Using a value of 50 instead of 100 for scene change threshold2 - does this give more or less weight to metric 2? i.e. does 50 make metric 2 more active in determining scene change vs. a value of 100?

Thanks
 

·
Registered
Joined
·
1,209 Posts
Using a value of 50 instead of 100 for scene change threshold2 - does this give more or less weight to metric 2? i.e. does 50 make metric 2 more active in determining scene change vs. a value of 100?

Thanks
The more you increase threshold2, the more metric2 will harm the detection of real scenes, because:

The 3rd number is now a *weighted* average. It's weighted like "(metric1 / threshold1 + metric2 / threshold2) / 2 * 10". Or to say it in words: The 3rd number now combines both metrics according to their thresholds, so that both have roughly equal weight. If the 3rd number exceeds 10.0, that's where a combined average metric would signal a scene change.
Therefore, 50 makes metric2 more active in determining a real scene change than 100.
 

·
Registered
Joined
·
3,005 Posts
Uuh, what's *that* you're talking about there!?!? :p

Edit: I'm really sorry, but I can't say much more about this yet. More info will come when I'm ready for that.

Edit2: Hopefully this thread will not be run over with speculation now, though?

@madshi Just wondering if you'll be selling stock. I'm in on the IPO. ;)
 

·
Registered
Joined
·
1,209 Posts
@madshi A sequence that could be interesting to test in the future is the one from BvS from 02:03:56 to 02:04:14.

I uploaded it here: https://www.mediafire.com/file/etdocknt9z99f52/muzzle_flashes.mkv/file

Disable substraction of previous frame OFF / ON







This sequence looks weird to me with "disable substraction of previous frame" OFF, the scene with Superman is staying dark several times after the bullets hit him.


@SOWK I think I prefer something like this for now:
 

·
Registered
JVC RS4500 | ST130 G4 135" | MRX 720 | MC303 MC152 | B&W 802D3, HTM1D3, 805D3, 702S2 | 4x15 IB Subs
Joined
·
8,444 Posts
Just looking at Neo's IT sample that he shared.

I've added the frame FALL for the cuts


1st metric / avg of both metrics / frame FALL

14.81 / 11.04
29.47 / 18.24 / 0.117
21.44 / 13.00 / 0.079
16.94 / 10.70 / 0.152
28.70 / 17.51 / 0.106
16.52 / 10.76 / 0.083
22.09 / 13.20 / 0.018
27.20 / 16.61 / 0.220
12.60 / 08.56 / 0.064
20.97 / 13.81 / 0.010
31.20 / 18.38 / 0.069
39.07 / 24.91 / 0.122
17.73 / 12.58 / 0.137
10.00 / 07.91 / 0.088



Those are very low frame FALL figures. The highest is 0.220

So I think you can work with this metric.






Yes. I'm finding them useful (although I was making a mental observation at the time and didn't record anything down to illustrate at hand).






I agree with this.



Assessing the qualitative state of a scene for a false positive/ missed scene change/cut can be tricky unless there's a glaring anomaly.

A higher scene threshold entry might prevent more false positives during a more volatile action scene i.e. Pacific Rim. However, those are relatively fast paced, blink and you miss it. Also, it's likely that the differential speeds will be in the higher threshold for those scenes and they'll have quicker entry speeds so they can readjust the dynamic target quickly. Also, when madVR develops the ability to look ahead, that might help with false positives in action scenes.

On the other hand, a higher scene threshold might not cause any discernible flaws in the picture from a failure to adjust the dynamic target accordingly due to a missed scene cut/change for a normal, longer, more stationary scene i.e. a missed dynamic adjustment target of e.g. 30 nits might not be too noticeable as long as the speed differentials are very low for that sort of scene.

I think we would need to look at many more scenes from various films.





I don't think it's helpful with higher FALL levels i.e. I was looking at a scene, which was causing a false positive with a frame FALL of 40.446

With 7 smooth iterations the hist stats (for the first and third column) were strikingly similar to the stats when the smooth iteration was disabled.


And reducing the stats for the hist might potentially narrow the threshold to work with on the lower end (but would need to confirm)





:D
@madshi has just posted something crazy exciting on doom and on his website!

It seems he just want to make "my/our" dream come true!

http://envy.madvr.com/


Basically madVR for all sources! So this means hdmi IN! :D:D:D
@madshi when is your baby coming? Can you tell us more?
Hopefully this standalone device can perform madVR's upscaling and external processing as well as HDR tone mapping. I filled out the survey :)
 

·
Registered
Joined
·
7,954 Posts
@madshi A sequence that could be interesting to test in the future is the one from BvS from 02:03:56 to 02:04:14.

sequence looks weird to me with "disable substraction of previous frame" OFF, the scene with Superman is staying dark several times after the bullets hit him.
We'll have to see how this looks once I start to "look ahead". Also for the "substraction of previous frame", we can try different strengths, or go back to a rolling average. But let's first finish metric2 tweaking. One step at a time... :)
 

·
Registered
Joined
·
1,209 Posts
Speaking of metric2, it harms the detection on HP and the Deathly Hallows: Part 2, from 01:31:25 to 01:36:37.

It doesn't catch any real scene and makes metric3 go lower than 10 sometimes, because it is so bad at detecting anything in this sequence.

Maybe this will give you an idea to improve the algo.

I uploaded a sample here if needed: https://www.mediafire.com/file/q32v5qodsg4u5o2/metric2_detection_test.mkv/file

Edit: I'm thinking, if we are going to stay on the detection subject for some time, what could help us to find problems and tweak more easily, would be to have a scene detection counter on the OSD for metric1/2/3, and the values of metric1/2/3 for the last detected scene(s), so we don't have to go frame by frame (which is time consuming).
 

·
Registered
Joined
·
7,954 Posts
Speaking of metric2, it harms the detection on HP and the Deathly Hallows: Part 2, from 01:31:25 to 01:36:37.

It doesn't catch any real scene and makes metric3 go lower than 10 sometimes, because it is so bad at detecting anything in this sequence.

Maybe this will give you an idea to improve the algo.

I uploaded a sample here if needed: https://www.mediafire.com/file/q32v5qodsg4u5o2/metric2_detection_test.mkv/file
Thanks for always providing a sample, that's really helpful!

Edit: I'm thinking, if we are going to stay on the detection subject for some time, what could help us to find problems and tweak more easily, would be to have a scene detection counter on the OSD for metric1/2/3, and the values of metric1/2/3 for the last detected scene(s), so we don't have to go frame by frame (which is time consuming).
I'm hoping it won't be too long. But who knows? In any case, your wish is my command! :)

--------------

So here's a new test build:

http://madshi.net/madVRhdrMeasure69.zip

The first option is the same as always (CLD New 2). The next 3 options are slight variations of CLD New 2. These variations are independent of each other. Which means every one which seems useful can probably be combined. So even if one of these look much better than the others, it's still worth testing all, because every small improvement can add up, if we combine these.

The next 3 options I'm not sure about. They could work, or maybe not. If you find in early testing that some of these don't work well, you can ignore those which don't appear to be useful, of course.
 

·
Registered
Joined
·
1,209 Posts
Results:



Doc: https://www.mediafire.com/file/74tl55d9k65zl16/Neo_V4.xlsx/file

Can you please verify the trial-and-error solver thresholds?

"CLD New 2 - prefer center" seems a bit better to catch real scenes with an overall low sum of misses, but all CLDs are pretty similar and the rendering times are the same.

Edit: I added more scenes, but it doesn't change anything:



Updated doc: http://www.mediafire.com/file/h94eba4tdwzrcci/Neo_V4.5.xlsx/file

I vote for "CLD New 2 - prefer center" for less misses on real scenes :)
 

·
Registered
Joined
·
7,954 Posts
Your Excel formulas and thresholds look alright to me. FWIW, I can modify the CLD New 2 "original" threshold so that it's very near to "prefer center". But why do you prefer false positive misses over real scene misses? Aren't both equally bad? Actually, I think image pumping (as bad as it is) is less evil than a sudden brightness jump in the middle of a scene, isn't it?

Anyway, to be honest I'm a bit disappointed in the results. I had hoped for a noticeable improvement, but I don't see it in the data, sadly.

One thing to consider: The "max adjustment" algo fixes the problem which CLD New 2 "original" still has, namely coming out of black, e.g. in the Lala Land scene you mentioned earlier. So I wonder if it's worth investigating into that direction more? The "max adjustment" currently limits APL adjustments to a max factor of 4.0 (or 0.25). Without that, the APL adjustment factor is completely unlimited. It being unlimited causes the problems with coming out of black. We could try different max factors. E.g. if unlimited works better than 4.0/0.25, we could maybe try 10.0/0.1?
 

·
Registered
Joined
·
1,479 Posts
Discussion Starter #5,520 (Edited)
But why do you prefer false positive misses over real scene misses? Aren't both equally bad? Actually, I think image pumping (as bad as it is) is less evil than a sudden brightness jump in the middle of a scene, isn't it?
Agreed. From what I observed, it's much worse to have a reset a of the target nits leading to a brightness jump within the same scene due to a false positive, than to miss a scene cut and play catch up.
 
5501 - 5520 of 9458 Posts
Top