AVS › AVS Forum › Display Devices › Digital Hi-End Projectors - $3,000+ USD MSRP › How to find the % APL of a particular scene?
New Posts  All Forums:Forum Nav:

How to find the % APL of a particular scene? - Page 2

post #31 of 91
1/6/07 at 2:24pm
Running that on the second VOB of Gladiator, and I get an AverageLuma of 25.92, or 10.16% (of 255).
Reply
Reply

AVS Top Picks

post #32 of 91
1/6/07 at 2:29pm
Thread Starter 
That looks pretty impressive, especially as it takes the info from the source rather than a screen grab that may not be accurate. I'd be interested in seeing what Erik thinks of it since he knows a lot more about it than I do!

Thanks.

Gary
Reply
Reply
post #33 of 91
1/6/07 at 3:24pm
Here's the first 1000 frames of gladiator, you can clearly see the fade-in of the Dreamworks logo, and then the beginning of the subsequent Universal logo.
LL
Reply
Reply
post #34 of 91
1/6/07 at 4:34pm
Quote:
Originally Posted by Gary Lightfoot View Post

That looks pretty impressive, especially as it takes the info from the source rather than a screen grab that may not be accurate.

Thanks, I was just reading this thread thinking "there's got to be a better way", and I knew people do all kinds of fancy stuff with AVISynth and figured it would be worth looking into.

So I finally got some time to play and it turned out (like I hoped) most everything we need is already there. I was extremely happy to see there was a handy way to write a file, if there hadn't been, it wouldn't have been nearly as cool.

Quote:
I'd be interested in seeing what Erik thinks of it since he knows a lot more about it than I do!

Me to (would like to get Erik's input). I know it's not absolutely correct in a rigorous sense, but I think the average of the luma channel should be close enough, and this method simple enough, and detailed enough, that it should be a good way to get a feel for what sort of levels real movies hold.

-edit

Oh, and one more thing I wanted to add. If this is more info than we need, ie if we just need a quick way to look at average luma of arbitrary scenes, then an easy way would be ffdshow, and then in the avisynth section, add the function call "ColorYUV(analyze=true)" which basically overlays the info my script extracts on the video.

That way you could just play various scenes and easily make note of the min/max/average luma values.
Reply
Reply
post #35 of 91
1/7/07 at 2:42am
Thread Starter 
In a nutshell, how would we get that set up onto our PCs? I don't have AVIsynth on there right now, but is it relatively easy to set up and use your script?

Gary
Reply
Reply
post #36 of 91
1/7/07 at 3:26am
Quote:
Originally Posted by Gary Lightfoot View Post

As mark said 'Wow'.
The other thing is that if scenes like that can have a lot of image depth, doesn't it disprove that high on/off CR displays create more depth of field in lower APL scenes and high ANSI ones in higher APL scenes, since this is a low APL scene with high depth of field when shown on a low (relatively speaking) contrast High ANSI projector (Sim2 C3X).
.

This Gladiator image is not really of low APL. If you would do a APL histogram of an average film this image would fall in the 25-50% quadrant, I'd guess (If anybody has actually analysed films in this regard please share your findings. This is a MOST interesting topic.). The result looks fine to me. Despite the DLP's considerably better ANSI the CRT can compete thanks to the high On-Off although this is NOT a challenging picture with really low APL where On-Off is expected to dominate. The result dispells the myth that ANSI of ~130 is really bad and DLPs blow CRTs away because of it unless you watch blackouts or something else extreme. Your normal film images in feature films are not of the ANSI checkerboard type and your average APL is rather low. As a result of this high end CRTs show good to superior depth with all kinds of images found in feature films while other technologies have weaknesses directly related to their On-Off. The higher the less weaknesses. No technology has superior depth for all kinds of images. That's in the future.
Edit: Seeing that efforts have started to analyse scenes and films it would be great if Eric's CR formula could be combined with the scripts that compute APL so we can get a histogram for intra frame contrast for whole films with typical CRT, LCD and DLP numbers. That would be very, very cool!
Reply
Reply
post #37 of 91
1/7/07 at 3:49am
Thread Starter 
Low ANSI on CRTs can produce things like halos and floating black levels where the black level changes colour and brightness with scene changes (especially on scope films shown on 16:9 screens), and especially in non optimal rooms. However, I would never have said DLPs blow CRT away (I leave that to the zealots or 'my projector is better than your projector' brigade). I've seen a few CRTs including G90s and Marquee 9500 Ultra, so I think I can say that that's definitely not the case. Other than absolute black level, I do think DLP produces a very good image though, although both do have a different 'look' to them. Pros and cons etc...

My main interest was the image depth perception idea that high on/off gives more low APL image depth, and high ANSI gives more high APL image depth. The scene I mentioned seems to prove otherwise, but the data that Erik has produced shows that overall, the numbers for both types of display give similar results and is probably why the scene looked so good on the C3X.

Like you though, I would have thought that the scene had a higher APL than that, but analyses seems to prove otherwise. The brightest part on my projector was around 54FL and the darkest part 0.05FL (measured at the screen facing the pj), but the majority of the image was in the darker range.

Gary
Reply
Reply
post #38 of 91
1/7/07 at 4:46am
Quote:
Originally Posted by Gary Lightfoot View Post

My main interest was the image depth perception idea that high on/off gives more low APL image depth, and high ANSI gives more high APL image depth. The scene I mentioned seems to prove otherwise, but the data that Erik has produced shows that overall, the numbers for both types of display give similar results and is probably why the scene looked so good on the C3X.
Gary

In general yes, high on/off gives more low APL image depth, and high ANSI gives more high APL image depth. The question is what is low and high, where is the middle area where both types of CR produce similar results and how much is actual luma distribution a factor (for same APL). Apparently this shot is in the middle area. Detailed analysis of whole films should make it clear which ranges are how common, how correlated APL and CR are (e.g. how relevant luma distribution variations within the same APL subset) and what results different technologies achieve in these ranges and with typical distributions. For example "Dark City" versus "Sahara".
Reply
Reply
post #39 of 91
1/7/07 at 4:54am
Thread Starter 
It would be interesting to see what kind of APL Sahara has and how good it would look on various types of display. I think it's a given that in the most part, Dark City would look better on a good CRT.

Gary
Reply
Reply
post #40 of 91
1/7/07 at 6:51am
Quote:
Originally Posted by Erik Garci View Post

My calculations are based on theory. To actually prove anything, I would measure it, not just calculate it.

What is the theoretical foundation of the scatter factor and formula for the checkerboard contrast?
Reply
Reply
post #41 of 91
1/7/07 at 7:17am
Quote:
Originally Posted by Gary Lightfoot View Post

In a nutshell, how would we get that set up onto our PCs? I don't have AVIsynth on there right now, but is it relatively easy to set up and use your script?

Gary


Basically you need AVISynth and DGIndex. Then you install AVISynth, and copy the dgdcode.dll from DGIndex into the AVISynth plugins directory.

After that's done, you open up a VOB from the DVD in DGIndex (or all of them), disable the audio demuxing (we don't need a copy of the audio), and choose save project to create a D2V file.

Then edit the Script I posted to point to the d2v file you just made. Then you should just be able to "run" the avisynth script by double clicking it, it should just play in WMP.
Reply
Reply
post #42 of 91
1/7/07 at 8:52am
Here's where you can download AVISynth:
http://www.videohelp.com/tools?tool=Avisynth

DGIndex is contained in the DGMPGDec tool:
http://www.videohelp.com/tools?tool=DGMPGDec

Both of these are legitimate freeware tools. Videohelp is a safe site to visit (as much as browsing the internet can be considered 'safe').
Reply
Reply
post #43 of 91
1/7/07 at 10:24am
Quote:
Originally Posted by stanger89 View Post

Here's the first 1000 frames of gladiator, you can clearly see the fade-in of the Dreamworks logo, and then the beginning of the subsequent Universal logo.

Really cool, can you quick explain what we are looking at, what are the "running" APL and Luma.

Can you compute somekind of average luma and APL for the 1000 frames.
Reply
Reply
post #44 of 91
1/7/07 at 1:46pm
Quote:
Originally Posted by ZoomAir View Post

Really cool, can you quick explain what we are looking at, what are the "running" APL and Luma.

That would be the average of every previous frame, ie the first frame, average of the first two, average of the first 3, etc.

Quote:


Can you compute somekind of average luma and APL for the 1000 frames.

That would be what the end of the Running APL/Luma graphs show.
Reply
Reply
post #45 of 91
1/7/07 at 2:24pm
OK, one more, here from the first ~24000 frames or 1000 seconds, I did a moving average of 24 frames so it's not quite so busy. Also attached is the raw source data if anyone wants to play with it.

Average Luma % (APL?) over that time is ~14.5%

There are a few really bright scenes in there up to ~60% APL, but overall it's under 25% I'd say.
LL

 

gladiator.zip 224.82421875k . file
Reply
Reply
post #46 of 91
1/7/07 at 2:33pm
Quote:
Originally Posted by stanger89 View Post

OK, one more, here from the first ~24000 frames or 1000 seconds, I did a moving average of 24 frames so it's not quite so busy.

Average Luma % (APL?) over that time is ~14.5%

There are a few really bright scenes in there up to ~60% APL, but overall it's under 25% I'd say.

Thanks, that is really cool to see, and shows that movies in fact is a dark medium, the more reason to find new contrast measurments as in the other thread.

Thanks again, i will try it myself.
Reply
Reply
post #47 of 91
1/8/07 at 12:42pm
Quote:
Originally Posted by mhafner View Post

What is the theoretical foundation of the scatter factor and formula for the checkerboard contrast?

When a projector produces light for a pixel on the screen, most of the light is contained within the pixel, but some of the light is scattered outside the pixel and thus falls onto other pixels. The scatter factor represents the portion of light that is scattered. To simplify the formulas, it is assumed that the light is scattered uniformly over the entire image. In reality, scattered light usually falls onto closer pixels more than farther pixels, so the calculations differ somewhat from actual measurements.

To calculate contrast, I perform several steps. I first calculate the average luminance of the projected image, without any scattered light. Then I calculate the amount of scattered light by multiplying the average luminance by the scatter factor. Then I add the scattered light to both the highest luminance and the lowest luminance in the image. Then I divide the highest luminance by the lowest luminance.
Reply
Reply
post #48 of 91
1/9/07 at 10:44am
Here is the new calculator.

Intrascene Contrast Calculator
Reply
Reply
post #49 of 91
1/9/07 at 11:13am
Quote:
Originally Posted by Erik Garci View Post

Here is the new calculator.

Intrascene Contrast Calculator

Wow! How does this work and what is it showing?
Reply
Reply
post #50 of 91
1/9/07 at 11:27am
Thread Starter 
Hi Erik,

That's a pretty impressive calculator.

On my 2650:1 pj I get around 1080:1 measured CR in my room from that Gladiator scene using a light meter on what look to me to be the lightest and darkest parts of the image (Gladiators stomach, and light through the trees from the top r/h area).

My room allows around 130:1 ANSI going on what I have measured, so I can get the reflectivity by working it backwards. Applying the reflectivity to the calculator, the numbers come out almost identical to the numbers I have physically measured (assuming a similar gamma, but it's still close enough even if the gamma was a little lower).

Brilliant!

Gary
Reply
Reply
post #51 of 91
1/9/07 at 11:36am
Quote:
Originally Posted by Mark Petersen View Post

Wow! How does this work and what is it showing?

You enter the on/off CR, ANSI CR, and gamma for each projector. You can also enter the room's reflectivity, or leave it at zero. Then click the "Calculate" button.

For each sample image, it calculates the average luma, or APL. It also calculates the average luminance, maximum luminance, minimum luminance, and the contrast ratio between the max and min.

You can click the name of the image to see the image.
Reply
Reply
post #52 of 91
1/9/07 at 2:21pm
Quote:
Originally Posted by Erik Garci View Post

You enter the on/off CR, ANSI CR, and gamma for each projector. You can also enter the room's reflectivity, or leave it at zero. Then click the "Calculate" button.

For each sample image, it calculates the average luma, or APL. It also calculates the average luminance, maximum luminance, minimum luminance, and the contrast ratio between the max and min.

You can click the name of the image to see the image.

Great work Erik! Looks great.

When you have a chance could you add perhaps two more scenes? One that is even darker than the darkest one, and one that is quite a bit brighter than the Gladiator one?

So it looks like ANSI CR kicks butt over the ON/OFF? The Sharp 20K is looking better and better. Just wish it had a few hundred more lumens @ D65...
Reply
Reply
post #53 of 91
1/9/07 at 2:30pm
It is interesting to see the significant impact that a higher gamma has on the results.

For instance at 2.2 there is a large gap in contrast between pj A and B. But when changing the gamma to 2.5 the pj with the higher on/off gains significant ground and closes the gap on B quite a bit.

So I take it then it is important to go with a higher gamma setting (particularly for dark movies) to get the most out of a pj with a high on/off rating?

This brings me to my next question Erik...

Apparently there are multiple ways to calculate gamma. Here is a description from the HCFR calibration software about their gamma measurement options:

Can you tell me which of these measurements types I'd be best off using to get my measured gamma equal to the method that your calculator assumes is being used for its gamma values? As a side note I believe ColorFacts uses the method equal to option #2 below.

Quote:
Originally Posted by Laric View Post

1) Display gamma

The gamma is compute in light to human eyes perception.
After approximation brightness (L) is expressed according to the signal (V) using the equation:
[1] L = V^gamma

Standards recommend to us a 2.5 Gamma target (2.45 after math) in this mode.

2) Display Gamma with Black Compensation (Default)

In practice one can never achieve an absolute black and to properly setup a projector (or screen) it is necessary to take in account of the black you can really obtain.

This is usually done by computing Gamma after removing the "absolute" black from every measure.

3) Camera Gamma (Standard Offset)

To have an accurate reproduction, some interpretations say that you must apply a symmetrical treatment to the encoding one. The reference equation being: [2] Y = ((V+offset)/(1+offset))^gamma

The "offset" is a parameter use at black encoding stage (and reproduction using this method). It's value is 0.099 on all video standard. (0.055 in sRGB)

4) Optimized (Regression)

This calculation method is based on the fact absolute black recorded by camera is the absolute black projector can produce.
But offset in standard is roughly for 200:1 contrast ratio, a lot lower than current ration projector can achieve.
The optimized method tries to find the Offset AND Gamma parameters that best match the measures, black included.

Pictures as contrasted as in method 2 can be achieved using this method with even a better black details level.
But the compute method -- and so the resulting projector settings -- are more dependant of proper and accurate low level probes readings.
Reply
Reply
post #54 of 91
1/9/07 at 2:33pm
Quote:
Originally Posted by lovingdvd View Post

So it looks like ANSI CR kicks butt over the ON/OFF? The Sharp 20K is looking better and better. Just wish it had a few hundred more lumens @ D65...

Really, I was just playing, and put in basically the RS1 (15000/350) and my IN76 (1500/550), and the theoretical RS1 stomps the theoretical IN76.
Reply
Reply
post #55 of 91
1/9/07 at 2:33pm
Quote:
Originally Posted by Erik Garci View Post

When a projector produces light for a pixel on the screen, most of the light is contained within the pixel, but some of the light is scattered outside the pixel and thus falls onto other pixels. The scatter factor represents the portion of light that is scattered. To simplify the formulas, it is assumed that the light is scattered uniformly over the entire image. In reality, scattered light usually falls onto closer pixels more than farther pixels, so the calculations differ somewhat from actual measurements.
To calculate contrast, I perform several steps. I first calculate the average luminance of the projected image, without any scattered light. Then I calculate the amount of scattered light by multiplying the average luminance by the scatter factor. Then I add the scattered light to both the highest luminance and the lowest luminance in the image. Then I divide the highest luminance by the lowest luminance.

Ok. Thanks. See Greg Roger's statements in the other thread. Your method should fail with any kind of not so evenly distributed picture, right? What do you think of my suggestion? Overkill?
Reply
Reply
post #56 of 91
1/9/07 at 2:36pm
Quote:
Originally Posted by stanger89 View Post

Really, I was just playing, and put in basically the RS1 (15000/350) and my IN76 (1500/550), and the theoretical RS1 stomps the theoretical IN76.

I was comparing based on two best of breed pjs - the RS1 and Sharp 20K. For the RS1 you need to use more like 12,000:1 with 270:1 ANSI CR. Perhaps the 12,000:1 may go up a bit based on throw and so forth but I'm not expecting more there. And by all accounts do not plan on ANSI CR being above 270:1. For the Sharp 20K you can use 7000:1 and 850:1 CR. You'll notice a much different result.
Reply
Reply
post #57 of 91
1/9/07 at 2:39pm
Quote:
Originally Posted by stanger89 View Post

Really, I was just playing, and put in basically the RS1 (15000/350) and my IN76 (1500/550), and the theoretical RS1 stomps the theoretical IN76.

The next step would be to automatically run whole films through the tool and analyse typical projectors over the whole film. Draw some nice curves, compute areas under the curves. Compute % of optimality.
Reply
Reply
post #58 of 91
1/9/07 at 2:42pm
Quote:
Originally Posted by lovingdvd View Post

I was comparing based on two best of breed pjs - the RS1 and Sharp 20K. For the RS1 you need to use more like 12,000:1 with 270:1 ANSI CR. Perhaps the 12,000:1 may go up a bit based on throw and so forth but I'm not expecting more there. And by all accounts do not plan on ANSI CR being above 270:1. For the Sharp 20K you can use 7000:1 and 850:1 CR. You'll notice a much different result.

Sharp better here makes sense, but use darker material and it changes. Use 30000:1 and it changes sooner. We need to run whole films to get a fair picture (pun intended).
Reply
Reply
post #59 of 91
1/9/07 at 2:44pm
Quote:
Originally Posted by lovingdvd View Post

I was comparing based on two best of breed pjs - the RS1 and Sharp 20K. For the RS1 you need to use more like 12,000:1 with 270:1 ANSI CR. Perhaps the 12,000:1 may go up a bit based on throw and so forth but I'm not expecting more there. And by all accounts do not plan on ANSI CR being above 270:1. For the Sharp 20K you can use 7000:1 and 850:1 CR. You'll notice a much different result.

True but if you are using 850:1 ANSI then you must also have some room gain, with 0,005 room gain which is a "good" room they are very similar in contrast in both scenes, however in a perfect room the Sharp pulls away on these scenes.

I use 15.000:1 and 250:1 for the JVC and 7000:1 850:1 for the Sharp.

Very cool calculator, if it is not much to ask, could you add a very dark scene like a star sky.

Again very nice work Eric, really appreciated
Reply
Reply
post #60 of 91
1/9/07 at 2:53pm
Quote:
Originally Posted by lovingdvd View Post

For the Sharp 20K you can use 7000:1 and 850:1 CR. You'll notice a much different result.

Well that kind of jives with what was posted (by Erik?) earlier, that a On-Off/ANSI ratio of 9 is about optimal, where one is holding the other back if it deviates from that.

For example, with my IN76, it's about 3x, and the higher On/Off makes a big difference, the low On/Off is holding the IN76 back more than the ANSI is benefiting it.

The RS1 is more like 45x, and when put up with something else, it's clear the ANSI is holding it back.

Looks like there are some interesting analysis' that can be done, for example:

Drop the "20k" On/Off to 5000 (5x), and the "RS1" regains it's lead, or if the RS1 somehow managed 400:1 ANSI, same result.

All in all, I think what we've found, or are finding, is very interesting and very valuable information in these threads.

1) It seems the measurements are validating the models.
2) 1a?) The models seem to correlate well with "real world" content/results.
3) The models allow easy, and apparently accurate, comparisons of projectors in ways we've never been able to quantify before.

I'm actually beginning to think that the really important thing that's been/being done here, is not the gathering of new data for projectors, but the investigation of how the current data (On/Off, ANSI) relates to real-world images and projectors.
Reply
Reply
New Posts  All Forums:Forum Nav:
  Return Home

AVS Top Picks

AVS › AVS Forum › Display Devices › Digital Hi-End Projectors - $3,000+ USD MSRP › How to find the % APL of a particular scene?