AVS Forum banner

5441 - 5460 of 9410 Posts

·
Registered
Joined
·
7,948 Posts
Just when I finished testing :rolleyes:
I'm sorry about that... :eek: I decided to investigate after your and Fer15's feedback, because something looked fishy to me, and then I found the bug.

On the positive side, I think the APL adjustment works much better now for metric2, and might just work for your IT scenes, too.

Can you check if there is a bug in the "adjust for brightness" option too? I got randomly low values on dark scenes with it enabled and consistent values with it disabled (same as the bug you fixed for the APL logic).
You talking metric1 here, right? I've just double checked: The code there is completely different and the same bug can't apply. The code looks alright to me. I don't think there's a bug in this feature. But if you have a good test scene which shows these random reactions, I guess I can have a deeper look.
 

·
Registered
Joined
·
1,208 Posts
You talking metric1 here, right? I've just double checked: The code there is completely different and the same bug can't apply. The code looks alright to me. I don't think there's a bug in this feature. But if you have a good test scene which shows these random reactions, I guess I can have a deeper look.
Yes, it's about metric1. I was testing today with scenes I know and the values got randomly low for apparently no reason.

For this sequence from It for instance: http://www.mediafire.com/file/vt4kavais8o9a0r/dark+scene.mkv

The values are as follow for metric1:

Adjust for brightness OFF "madVRhdrMeasure53" / Adjust for brightness ON "madVRhdrMeasure53"

68.23 / 29.83
27.43 / 22.04
17.13 / 17.13
31.51 / 29.02
17.03 / 17.03
23.32 / 22.64
27.48 / 27.48
44.36 / 12.96
36.26 / 21.21
31.74 / 31.64
39.42 / 39.42
21.69 / 18.19
16.27 / 10.63
 

·
Registered
Joined
·
7,948 Posts
Yes, it's about metric1. I was testing today with scenes I know and the values got randomly low for apparently no reason.

For this sequence from It for instance: http://www.mediafire.com/file/vt4kavais8o9a0r/dark+scene.mkv

The values are as follow for metric1:

Adjust for brightness OFF "madVRhdrMeasure53" / Adjust for brightness ON "madVRhdrMeasure53"

68.23 / 29.83
27.43 / 22.04
17.13 / 17.13
31.51 / 29.02
17.03 / 17.03
23.32 / 22.64
27.48 / 27.48
44.36 / 12.96
36.26 / 21.21
31.74 / 31.64
39.42 / 39.42
21.69 / 18.19
16.27 / 10.63
Which frame numbers are these numbers from in the sample?
 

·
Registered
Joined
·
7,948 Posts
If you have weird results (e.g. really bad numbers for "always adjust"), too, you could try this build instead:

http://madshi.net/madVRhdrMeasure63.zip

The numbers for "always ignore" are the same in builds 62 and 63, so no need to retest them. But the numbers for "always adjust" and "choose lowest diff" are different in builds 62 and 63.

(I'm sorry for throwing so many test builds at you... :( At least the number of options to test is low this time.)

Edit: Fer15 confirms build 63 seems to produce *much* better results for "always adjust" and "choose lowest diff" compared to build 62. So trash build 62, please.
 

·
Registered
Joined
·
466 Posts
Not sure if you still need any further test material. Arrival from 1:01:52 to 1:02:30 has several visible (3 or 4) brightness adjustments. Using the latest 63. Indoor low light scene. Conversation between the two principles. I suspect the seen detection is not catching all it should? Arrival is often pegged to my "real nits" value (90) and would like to go lower I suspect as the reported source nits is 20's and 30's often. Not sure if that is the reason it seems to show more fluctuations than other movies. I watched Widows the other night (a few versions ago) and only saw a few times that looked a little funky with a brightness shift so pretty solid overall.
 

·
Registered
Joined
·
7,948 Posts
@Neo-XP, thanks! Interestingly, Fer15 has somewhat different results once more. For him, "Choose lowest diff" comes out on top. Now if I replace your "real scenes" with the data from Fer15, the combined results have "Choose lowest diff" and "Always adjust" almost with identical results (different numbers, but same ratio). In any case, it seems as if "Always adjust" is the safest option for now, because it does well both in your and Fer15's tests.

Btw, I think the ratio (real scenes / false positives) is a better metric than the difference (real scenes - false positives).
 

·
Registered
Joined
·
1,208 Posts
Not sure if you still need any further test material. Arrival from 1:01:52 to 1:02:30 has several visible (3 or 4) brightness adjustments. Using the latest 63. Indoor low light scene. Conversation between the two principles. I suspect the seen detection is not catching all it should? Arrival is often pegged to my "real nits" value (90) and would like to go lower I suspect as the reported source nits is 20's and 30's often. Not sure if that is the reason it seems to show more fluctuations than other movies. I watched Widows the other night (a few versions ago) and only saw a few times that looked a little funky with a brightness shift so pretty solid overall.
Thanks, I checked Arrival and it seems that the scenes detection is not triggering even if metric3 is over 10, that's weird.

New test build:

http://madshi.net/madVRhdrMeasure64.zip

This has the same old "always adjust" and "choose lowest diff" as the previous build, but it adds one new "always adjust" and two new "choose lowest diff" variants. So there's 5 options overall, but for 2 of them you can simply copy the data from build 63.
I will wait a few hours before testing this :D

Is it possible to add the "adjust for brightness" on/off trigger, so I can test everything at the same time and add the values to the table?

Thanks!
 

·
Registered
Joined
·
7,948 Posts
Thanks, I checked Arrival and it seems that the scenes detection is not triggering even if metric3 is over 10, that's weird.
Is it just Arrival, or does the metric3 > 10.0 not work with any movie/scene?

I will wait a few hours before testing this :D
Hehe... :D

Is it possible to add the "adjust for brightness" on/off trigger, so I can test everything at the same time and add the values to the table?
Ok, will do.
 

·
Registered
Joined
·
466 Posts
Is it just Arrival, or does the metric3 > 10.0 not work with any movie/scene?
Hopefully not too dumb a question but where does it display the values like metric 3 and the others used for the sheet results. I might be of more help if I can review/provide this.

Also interesting is I went and watched the same sequence with my tweaked measurement file and it had similar if not worse brightness shifting. I will have to go back now and watch on my TV with no tone mapping just to make sure the source material is not wonky (beyond very low nit) but I don't think it is. Maybe its low nits skews something in the "fall" adjustment algorithm? Maybe I will post that sequence in that thread and see if any info comes from it. EDIT: I will also retest with latest. Missed that somehow.
 

·
Registered
Joined
·
1,208 Posts
Hopefully not too dumb a question but where does it display the values like metric 3 and the others used for the sheet results. I might be of more help if I can review/provide this.
They are displayed on the lum graph at the top-right corner of the OSD:



They are in order: metric1 is only based on luminance distribution changes (correct me if I'm wrong), metric2 is the scene detection we are testing now (a more complex one), and metric3 is just the average of metric1 and metric2.

Also, the scene detection was not working properly in "madVRhdrMeasure64", but seems fixed in "madVRhdrMeasure65" (I just checked).
 

·
Registered
Joined
·
7,948 Posts
Fer15 also had very high results in some scenes, especially IT, but all movie scenes combined, the real scene vs false positive ratio actually worked out pretty well for "always adjust - new" in Fer15's tests. It wasn't the best option in his tests, but also not the worst. I *think* the code is ok, but of course there's always a chance I missed something.

One thing I'm wondering about is if maybe the metric2 FALL adjustment for dark scenes which I did a while ago is now superfluous or even counter productive, using the latest metric2 improvements? Maybe that's why the numbers are so high now for IT when using "always adjust - new"? Just a wild guess, though...
 

·
Registered
Joined
·
1,208 Posts
One thing I'm wondering about is if maybe the metric2 FALL adjustment for dark scenes which I did a while ago is now superfluous or even counter productive, using the latest metric2 improvements? Maybe that's why the numbers are so high now for IT when using "always adjust - new"? Just a wild guess, though...

The other options are working well with consistent values, it's just the "always adjust - new" which has this problem.


Do I start testing like this? I will ignore "always adjust - new" then.
 

·
Registered
Joined
·
7,948 Posts
I could add an option to disable the FALL tweak for IT, maybe that helps avoiding those large numbers? But Fer15's measurements say that "always adjust - new" measures better for him than "always adjust - old" - even though IT produces such high numbers.
 

·
Registered
Joined
·
1,208 Posts
I could add an option to disable the FALL tweak for IT, maybe that helps avoiding those large numbers? But Fer15's measurements say that "always adjust - new" measures better for him than "always adjust - old" - even though IT produces such high numbers.

That does not make sense, I got lower values on real cuts than on a calm scene when the FALL is low, but only with this one.


Maybe an option to disable the FALL tweak could work with this new algo.
 
5441 - 5460 of 9410 Posts
Top