AVS APL Study - Adjunct to AVS Contrast Project - Page 6 - AVS Forum | Home Theater Discussions And Reviews
Baselworld is only a few weeks away. Getting the latest news is easy, Click Here for info on how to join the Watchuseek.com newsletter list. Follow our team for updates featuring event coverage, new product unveilings, watch industry news & more!


Forum Jump: 
 22Likes
Reply
Thread Tools
post #151 of 177 Unread 09-26-2015, 08:40 PM
AVS Addicted Member
 
Stereodude's Avatar
 
Join Date: Jan 2002
Location: Detroit Metro Area
Posts: 12,856
Mentioned: 23 Post(s)
Tagged: 0 Thread(s)
Quoted: 2347 Post(s)
Liked: 1507
BTW, I didn't realize that Guardians of the Galaxy had both 16:9 and 2.35:1 content. I ran it cropped. I will run it again without cropping.

I'll also see about the clip and images you provided.

Edit: So only the 3D changes AR. The 2D is 2.40 all the way though. I'm not really sure how to treat them. Do you want numbers for the 2D release with the top and bottom black bars cut off?

Last edited by Stereodude; 09-27-2015 at 04:55 PM.
Stereodude is offline  
Sponsored Links
Advertisement
 
post #152 of 177 Unread 09-26-2015, 10:07 PM
AVS Addicted Member
 
Stereodude's Avatar
 
Join Date: Jan 2002
Location: Detroit Metro Area
Posts: 12,856
Mentioned: 23 Post(s)
Tagged: 0 Thread(s)
Quoted: 2347 Post(s)
Liked: 1507
Quote:
Originally Posted by darinp2 View Post
While you are engaged in this I would appreciate it if you could help with a sanity check, just to get more confidence that we are analyzing these correctly.

I cut a 1 second clip out of North By Northwest from the brightest part I found and put it here:

https://www.dropbox.com/s/lub5hchysa...t1sec.avi?dl=0
I ran every frame and got:

max: 28.618
mean: 24.432
median: 24.038

Zoyd's script attempts to emulate BT.1886. Converting for a gamma of 2.2 I got:

max: 29.894
mean: 25.302
median: 25.300

Last edited by Stereodude; 09-27-2015 at 10:38 AM. Reason: correct number for 2.2 gamma again
Stereodude is offline  
post #153 of 177 Unread 09-27-2015, 05:28 AM
AVS Addicted Member
 
Stereodude's Avatar
 
Join Date: Jan 2002
Location: Detroit Metro Area
Posts: 12,856
Mentioned: 23 Post(s)
Tagged: 0 Thread(s)
Quoted: 2347 Post(s)
Liked: 1507
@zoyd Are you sure the linearization line is correct?

Code:
#dither input for better precision at low levels
#RGB on 3 stacked 16 bit interleaved frames
dither_convert_yuv_to_rgb(matrix="709", mode=6, tv_range=true, output="rgb48y")

#srgb curve with gcor factor to approximate BT.1886
Dither_y_gamma_to_linear(curve="srgb", gcor=1.044, tv_range_in=false, tv_range_out=false)
Your color space converting from BT.709 to sRGB with dither_convert_yuv_to_rgb and are then linearizing the luma channel with Dither_y_gamma_to_linear. Unless the color space conversion changes also change the gamma curve, shouldn't you use 709 for the linearization? I didn't see anything in the dither documentation that leads me to believe the gamma curve is adjusted in dither_convert_yuv_to_rgb. Thinking alone those lines, if it did change the gamma to sRGB automatically I would expect that Dither_y_gamma_to_linear wouldn't need a curve argument because it would always get an sRGB gamma curve.

Last edited by Stereodude; 09-27-2015 at 05:32 AM.
Stereodude is offline  
post #154 of 177 Unread 09-27-2015, 07:00 AM
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 6,049
Mentioned: 73 Post(s)
Tagged: 0 Thread(s)
Quoted: 1131 Post(s)
Liked: 886
dither_convert_yuv_to_rgb doesn't touch gamma, it merely converts YUV to RGB using the rec. 709 matrix.

Dither_y_gamma_to_linear linearizes each RGB channel using the the srgb gamma curve which is close to BT.1886 with a contrast ratio of 1176 and has an average gamma close to 2.2. I add a little power to that curve (gcor = 1.044) so that it approximates a BT.1886 curve with a contrast ratio of 2500 and average gamma of 2.3. If you used the rec.709 curve you would get an average gamma of 1.96.

Last edited by zoyd; 09-28-2015 at 05:40 AM.
zoyd is online now  
post #155 of 177 Unread 09-27-2015, 07:31 AM
Senior Member
 
Soulnight's Avatar
 
Join Date: Dec 2011
Posts: 210
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 107 Post(s)
Liked: 77
Thank you Stereo Dude for also contributing to this thread.

I would suggest that we all use a gamma of 2.2 and not the BT 1886 to stay consistent.

Soulnight is offline  
post #156 of 177 Unread 09-27-2015, 09:28 AM
AVS Addicted Member
 
Stereodude's Avatar
 
Join Date: Jan 2002
Location: Detroit Metro Area
Posts: 12,856
Mentioned: 23 Post(s)
Tagged: 0 Thread(s)
Quoted: 2347 Post(s)
Liked: 1507
Quote:
Originally Posted by zoyd View Post
dither_convert_yuv_to_rgb doesn't touch gamma, it merely converts YUV to RGB using the rec. 709 matrix.

Dither_y_gamma_to_linear linearizes each RGB channel using the the srgb gamma curve which is close to BT.1886 with black level of 0.1 cd/m^2 and has an average gamma close to 2.2. I add a little power to that curve (gcor = 1.044) so that it approximates a BT.1886 curve with 0.05 cd/m^2 with an average gamma of 2.3. If you used the rec.709 curve you would get an average gamma of 1.96.
So to use a traditional display gamma of 2.2 I should use 709 in Dither_y_gamma_to_linear without the gcor parameter correct?
Stereodude is offline  
post #157 of 177 Unread 09-27-2015, 09:40 AM
AVS Addicted Member
 
Stereodude's Avatar
 
Join Date: Jan 2002
Location: Detroit Metro Area
Posts: 12,856
Mentioned: 23 Post(s)
Tagged: 0 Thread(s)
Quoted: 2347 Post(s)
Liked: 1507
Quote:
Originally Posted by Soulnight View Post
I would suggest that we all use a gamma of 2.2 and not the BT 1886 to stay consistent.
Well what to use, and if it should be 2.2 gets into a much larger discussion. Mainly driven by the fact that CRTs didn't really have a gamma of 2.2. So the idea that the CRT reference monitors that all content was graded on had a gamma of 2.2 isn't really correct. They're more like 2.35.

That said, BT.709 2.2 vs. Zoyd's BT.1886 curve does have noticeable difference, at least in brighter content. I updated post 152 and you can see the difference between the two on Darin's North by Northwest clip.
Stereodude is offline  
post #158 of 177 Unread 09-27-2015, 10:03 AM
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 6,049
Mentioned: 73 Post(s)
Tagged: 0 Thread(s)
Quoted: 1131 Post(s)
Liked: 886
Quote:
Originally Posted by Stereodude View Post
So to use a traditional display gamma of 2.2 I should use 709 in Dither_y_gamma_to_linear without the gcor parameter correct?
No, that will give you an average 1.96 gamma response. Use either sRGB without gcor, or rec.709 with gcor = 1.2
zoyd is online now  
post #159 of 177 Unread 09-27-2015, 10:10 AM
AVS Addicted Member
 
Stereodude's Avatar
 
Join Date: Jan 2002
Location: Detroit Metro Area
Posts: 12,856
Mentioned: 23 Post(s)
Tagged: 0 Thread(s)
Quoted: 2347 Post(s)
Liked: 1507
Quote:
Originally Posted by zoyd View Post
No, that will give you an average 1.96 gamma response. Use either sRGB without gcor, or rec.709 with gcor = 1.2
Okay, I don't exactly understand why that is, but I'll defer to your expertise.
Stereodude is offline  
post #160 of 177 Unread 09-27-2015, 10:31 AM
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 6,049
Mentioned: 73 Post(s)
Tagged: 0 Thread(s)
Quoted: 1131 Post(s)
Liked: 886
Quote:
Originally Posted by Stereodude View Post
Okay, I don't exactly understand why that is, but I'll defer to your expertise.
Because the Rec.709 curve is not a display referenced transfer function, it's an encoding function that is quite a bit shallower (lower gamma) than sRGB even though it uses an exponent of 2.22. sRGB is a display referenced transfer function with an average power type exponent closer to 2.2

zoyd is online now  
post #161 of 177 Unread 09-27-2015, 10:40 AM
AVS Addicted Member
 
Stereodude's Avatar
 
Join Date: Jan 2002
Location: Detroit Metro Area
Posts: 12,856
Mentioned: 23 Post(s)
Tagged: 0 Thread(s)
Quoted: 2347 Post(s)
Liked: 1507
Quote:
Originally Posted by zoyd View Post
Use either sRGB without gcor, or rec.709 with gcor = 1.2
FWIW, these two options don't give the same results.

The former gives higher numbers than your BT.1886 approximation and the latter gives numbers lower.

Quote:
Originally Posted by zoyd View Post
Because the Rec.709 curve is not a display referenced transfer function, it's an encoding function that is quite a bit shallower (lower gamma) than sRGB even though it uses an exponent of 2.22. sRGB is a display referenced transfer function with an average power type exponent closer to 2.2

Okay, thanks for the explanation.
Stereodude is offline  
post #162 of 177 Unread 09-27-2015, 11:17 AM
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 6,049
Mentioned: 73 Post(s)
Tagged: 0 Thread(s)
Quoted: 1131 Post(s)
Liked: 886
Quote:
Originally Posted by Stereodude View Post
FWIW, these two options don't give the same results.

The former gives higher numbers than your BT.1886 approximation...
That's expected.


Quote:
...and the latter gives numbers lower.
That's not, just use sRGB then.
zoyd is online now  
post #163 of 177 Unread 09-27-2015, 07:53 PM
AVS Addicted Member
 
darinp2's Avatar
 
Join Date: Sep 2003
Location: Seattle, WA
Posts: 23,085
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 890 Post(s)
Liked: 1008
Quote:
Originally Posted by zoyd View Post
Dither_y_gamma_to_linear linearizes each RGB channel using the the srgb gamma curve which is close to BT.1886 with black level of 0.1 cd/m^2 ...
What white level does that 0.1 go with?
Quote:
Originally Posted by Stereodude View Post
Well what to use, and if it should be 2.2 gets into a much larger discussion. Mainly driven by the fact that CRTs didn't really have a gamma of 2.2. So the idea that the CRT reference monitors that all content was graded on had a gamma of 2.2 isn't really correct. They're more like 2.35.
Yep. I stuck with the 2.2 in the original files not because I considered it correct, but because I figured it was a reasonable compromise from having to assume a contrast ratio and should error on the side of brighter compared to something like BT.1886 assuming perfect CR (zero black level). In my presentation at SID 2015 I was also trying to avoid having the subject jump off into the "correct" gamma when my point was largely how far real movie content tends to be from the 50% ADL of the ANSI checkerboards, and that holds true at just about any reasonable gamma. I did mention that results would be lower with 2.4 gamma in my latest document.

I'm still stuck at 8 bit without dithering. For the ~1 second clip from North By Northwest I used the following avs file:
Code:
LoadPlugin("C:\ffms2\ffms2-2.20-icl\x86\ffms2.dll")
clip = FFVideoSource("C:\Video\NBNW\brightPics\nbnwBright1sec.avi", cachefile = "C:\Video\NBNW\brightPics\nbnwBright1sec temp files\nbnwBright1sec.ffindex", fpsnum = 24000, fpsden = 1001)
filename = "nbnwBrightSec.everyFrame.output.txt"
convertToRGB(clip,"Rec709", chromaresample="lanczos")
clip = Levels(0, 1/2.2, 255, 0, 255, coring=false, dither=false)
ConvertToYV16(clip, false, "Rec709")
#Crop(32,32,-32,-32)
Selectevery(1,0)
AssumeFps(240)
colon = " : "
WriteFile(filename, "current_frame", "colon", "YPlaneMin", "colon", "YPlaneMax", "colon", "AverageLuma")
WriteFileStart(filename, """ "AverageLuma for NBNW with level fix" """)
WriteFileStart(filename, """ "Frame : MinLuma : MaxLuma : AverageLuma" """ )
WriteFileEnd(filename, """ "End" """)
ConvertToYUY2()
and got these results for each frame.
Code:
Frame : MinLuma : MaxLuma : AverageLuma
0 : 16 : 213 : 72.591393
1 : 16 : 213 : 72.591393
2 : 16 : 207 : 72.537430
3 : 16 : 207 : 72.537430
4 : 16 : 211 : 73.983269
5 : 16 : 212 : 78.012131
6 : 16 : 212 : 78.012131
7 : 16 : 221 : 82.745598
8 : 16 : 213 : 80.597092
9 : 16 : 212 : 78.915733
10 : 16 : 210 : 77.030670
11 : 16 : 212 : 76.302277
12 : 16 : 212 : 76.302277
13 : 16 : 211 : 74.430405
14 : 16 : 212 : 72.339638
15 : 16 : 207 : 71.194717
16 : 16 : 207 : 70.286270
17 : 16 : 216 : 70.161880
18 : 16 : 208 : 69.219711
19 : 16 : 209 : 68.439796
20 : 16 : 207 : 67.047089
21 : 16 : 208 : 66.626381
22 : 16 : 211 : 66.051247
23 : 16 : 213 : 65.536591
24 : 16 : 213 : 65.536591
25 : 16 : 217 : 64.583511
End
So, peak of 82.75 becoming (82.75-16)/219=30.5% ADL.


Thanks,
Darin

This is the AV Science Forum. Please don't be gullible and please do remember the saying, "Fool me once, shame on you. Fool me twice, shame on me."

Last edited by darinp2; 09-27-2015 at 08:00 PM.
darinp2 is offline  
post #164 of 177 Unread 09-27-2015, 09:01 PM
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 6,049
Mentioned: 73 Post(s)
Tagged: 0 Thread(s)
Quoted: 1131 Post(s)
Liked: 886
Quote:
Originally Posted by darinp2 View Post
What white level does that 0.1 go with?
The best fit Bt.1886 curve to sRGB has a contrast ratio of 1176 so that would yield a white level of 117.6 cd/m^2.

Last edited by zoyd; 09-28-2015 at 04:57 AM.
zoyd is online now  
post #165 of 177 Unread 09-27-2015, 09:27 PM
AVS Addicted Member
 
darinp2's Avatar
 
Join Date: Sep 2003
Location: Seattle, WA
Posts: 23,085
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 890 Post(s)
Liked: 1008
Quote:
Originally Posted by zoyd View Post
The BT.1886 shape doesn't depend on white level as strongly as black level ...
I thought it was based on contrast ratio, meaning the exact same shape with double the white and black levels, or half the white and black levels for instance (any case with the same ratio). In this case same shape meaning the same ratios between every level.

Is that not the case?

Did you end up plugging 0.1 in somewhere when running the analysis? If so I would have expected it to need a white level too.

Thanks,
Darin

Last edited by darinp2; 09-27-2015 at 09:30 PM.
darinp2 is offline  
post #166 of 177 Unread 09-27-2015, 09:37 PM
AVS Addicted Member
 
Stereodude's Avatar
 
Join Date: Jan 2002
Location: Detroit Metro Area
Posts: 12,856
Mentioned: 23 Post(s)
Tagged: 0 Thread(s)
Quoted: 2347 Post(s)
Liked: 1507
Quote:
Originally Posted by darinp2 View Post
So, peak of 82.75 becoming (82.75-16)/219=30.5% ADL.
Very close to my 2.2 gamma results
Code:
Luma/Luminance statistics for NBNW Sample 09/27/2015 12:35:43
Color space: 1 Color range: 1 Reported FPS: 23.976025
Frame , Current time, Fractional minutes, Hour, Minute, Seconds, Ystim[%], U, V, Ystim accum.[%], U accum., V accum., MinLuminance , MaxLuminance , AverageLuminance, MedianLuminance, Average Luminance[%]
0 , 12:35:55 , 0.000000 , 0 , 0 , 0.000000 , 50.104782 , 131.191956 , 121.574028 , 50.104782 , 131.191956 , 121.574028 , 16 , 212 , 71.435532 , 84 , 25.313028
1 , 12:35:55 , 0.000683 , 0 , 0 , 0.041000 , 50.104782 , 131.191956 , 121.574028 , 50.104782 , 131.191956 , 121.574028 , 16 , 212 , 71.435532 , 84 , 25.313028
2 , 12:35:56 , 0.001383 , 0 , 0 , 0.083000 , 50.086933 , 131.100601 , 121.669014 , 50.098831 , 131.161499 , 121.605690 , 16 , 206 , 71.380028 , 84 , 25.287683
3 , 12:35:57 , 0.002083 , 0 , 0 , 0.125000 , 50.339352 , 130.602539 , 122.089111 , 50.158962 , 131.021759 , 121.726547 , 16 , 208 , 71.727837 , 84 , 25.446501
4 , 12:35:57 , 0.002767 , 0 , 0 , 0.166000 , 50.832291 , 129.933502 , 122.970375 , 50.293629 , 130.804108 , 121.975311 , 16 , 210 , 72.796593 , 85 , 25.934517
5 , 12:35:58 , 0.003467 , 0 , 0 , 0.208000 , 52.457855 , 129.037659 , 124.021370 , 50.654331 , 130.509705 , 122.316322 , 16 , 210 , 76.770546 , 86 , 27.749107
6 , 12:35:58 , 0.004167 , 0 , 0 , 0.250000 , 53.978634 , 128.311920 , 124.741501 , 51.129234 , 130.195740 , 122.662773 , 16 , 212 , 80.872803 , 87 , 29.622284
7 , 12:35:58 , 0.004850 , 0 , 0 , 0.291000 , 54.170845 , 126.704819 , 125.468742 , 51.509434 , 129.759369 , 123.013519 , 16 , 221 , 81.467560 , 88 , 29.893862
8 , 12:35:59 , 0.005550 , 0 , 0 , 0.333000 , 53.310486 , 125.771812 , 125.988045 , 51.709553 , 129.316299 , 123.344025 , 16 , 212 , 79.347931 , 86 , 28.925995
9 , 12:35:59 , 0.006250 , 0 , 0 , 0.375000 , 52.646019 , 125.084541 , 126.418175 , 51.803196 , 128.893127 , 123.651436 , 16 , 210 , 77.686340 , 86 , 28.167278
10 , 12:35:59 , 0.006950 , 0 , 0 , 0.417000 , 51.825760 , 125.292747 , 126.516106 , 51.805244 , 128.565826 , 123.911865 , 16 , 208 , 75.835289 , 85 , 27.322048
11 , 12:35:59 , 0.007633 , 0 , 0 , 0.458000 , 51.440826 , 125.397636 , 126.395592 , 51.774883 , 128.301804 , 124.118843 , 16 , 210 , 75.122101 , 86 , 26.996393
12 , 12:35:59 , 0.008333 , 0 , 0 , 0.500000 , 50.620316 , 125.085167 , 127.016685 , 51.686066 , 128.054367 , 124.341751 , 16 , 209 , 73.286392 , 85 , 26.158169
13 , 12:35:59 , 0.009033 , 0 , 0 , 0.542000 , 50.126534 , 125.460808 , 127.130074 , 51.574673 , 127.869118 , 124.540916 , 16 , 208 , 72.326561 , 84 , 25.719891
14 , 12:35:59 , 0.009717 , 0 , 0 , 0.583000 , 49.576641 , 125.524551 , 127.315857 , 51.441475 , 127.712814 , 124.725914 , 16 , 210 , 71.250687 , 83 , 25.228622
15 , 12:35:59 , 0.010417 , 0 , 0 , 0.625000 , 49.002735 , 124.857948 , 127.752579 , 51.289047 , 127.534386 , 124.915077 , 16 , 205 , 70.134666 , 83 , 24.719025
16 , 12:36:00 , 0.011117 , 0 , 0 , 0.667000 , 48.499043 , 124.957359 , 128.038681 , 51.124931 , 127.382797 , 125.098824 , 16 , 206 , 69.255386 , 81 , 24.317528
17 , 12:36:00 , 0.011817 , 0 , 0 , 0.709000 , 48.299149 , 125.170372 , 127.720329 , 50.967941 , 127.259880 , 125.244461 , 16 , 214 , 69.142044 , 82 , 24.265774
18 , 12:36:00 , 0.012500 , 0 , 0 , 0.750000 , 47.777138 , 124.625679 , 128.103271 , 50.800007 , 127.121239 , 125.394920 , 16 , 206 , 68.229782 , 81 , 23.849215
19 , 12:36:00 , 0.013200 , 0 , 0 , 0.792000 , 47.327599 , 124.699783 , 127.995316 , 50.626385 , 127.000168 , 125.524940 , 16 , 207 , 67.474045 , 80 , 23.504129
20 , 12:36:00 , 0.013900 , 0 , 0 , 0.834000 , 46.604763 , 124.224930 , 128.195587 , 50.434879 , 126.868011 , 125.652115 , 16 , 205 , 66.122559 , 79 , 22.887012
21 , 12:36:00 , 0.014583 , 0 , 0 , 0.875000 , 46.302910 , 123.821114 , 128.711685 , 50.247063 , 126.729515 , 125.791191 , 16 , 206 , 65.717766 , 79 , 22.702175
22 , 12:36:00 , 0.015283 , 0 , 0 , 0.917000 , 45.907764 , 124.208176 , 129.020874 , 50.058395 , 126.619896 , 125.931610 , 16 , 210 , 65.167786 , 78 , 22.451042
23 , 12:36:00 , 0.015983 , 0 , 0 , 0.959000 , 45.556004 , 123.372276 , 129.656906 , 49.870796 , 126.484573 , 126.086830 , 16 , 212 , 64.671974 , 77 , 22.224646
24 , 12:36:00 , 0.016683 , 0 , 0 , 1.001000 , 45.219036 , 123.528091 , 129.640564 , 49.684727 , 126.366318 , 126.228981 , 16 , 208 , 64.257507 , 76 , 22.035391
25 , 12:36:00 , 0.017367 , 0 , 0 , 1.042000 , 44.799477 , 124.209038 , 129.171280 , 49.496834 , 126.283348 , 126.342148 , 16 , 215 , 63.758419 , 76 , 21.807497
End
Stereodude is offline  
post #167 of 177 Unread 09-27-2015, 11:59 PM
AVS Addicted Member
 
darinp2's Avatar
 
Join Date: Sep 2003
Location: Seattle, WA
Posts: 23,085
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 890 Post(s)
Liked: 1008
Quote:
Originally Posted by Stereodude View Post
Very close to my 2.2 gamma results
Thanks. Pretty close for that bright stuff.

I expect that the 8 bit I am using would show its limitations in the darkest scenes. I put up a dark scene of the type that INFOCOMM claims requires 80:1 system ANSI CR to be able to see what is going on:

https://www.dropbox.com/s/prdxhy2ub7...oods2.avi?dl=0

Here is what I got with the same kind of script and 2.2 gamma:
Code:
Frame : MinLuma : MaxLuma : AverageLuma
0 : 16 : 212 : 16.338640
1 : 16 : 212 : 16.338640
2 : 16 : 221 : 16.341902
3 : 16 : 221 : 16.341902
4 : 16 : 220 : 16.343262
5 : 16 : 218 : 16.341234
6 : 16 : 218 : 16.341234
7 : 16 : 90 : 16.333231
8 : 16 : 211 : 16.334780
9 : 16 : 220 : 16.342569
10 : 16 : 222 : 16.343973
11 : 16 : 223 : 16.343220
12 : 16 : 223 : 16.343220
13 : 16 : 222 : 16.344625
14 : 16 : 223 : 16.344189
15 : 16 : 220 : 16.342085
16 : 16 : 221 : 16.341789
17 : 16 : 222 : 16.342525
18 : 16 : 224 : 16.340569
19 : 16 : 224 : 16.341648
20 : 16 : 223 : 16.342867
21 : 16 : 222 : 16.341356
22 : 16 : 221 : 16.340574
23 : 16 : 218 : 16.333244
I'm guessing that 16 bit would calculating higher for a scene like this, but I'm not sure how much higher.

--Darin
Ericglo likes this.

This is the AV Science Forum. Please don't be gullible and please do remember the saying, "Fool me once, shame on you. Fool me twice, shame on me."

Last edited by darinp2; 09-28-2015 at 12:03 AM.
darinp2 is offline  
post #168 of 177 Unread 09-28-2015, 04:56 AM
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 6,049
Mentioned: 73 Post(s)
Tagged: 0 Thread(s)
Quoted: 1131 Post(s)
Liked: 886
Quote:
Originally Posted by darinp2 View Post
I thought it was based on contrast ratio, meaning the exact same shape with double the white and black levels, or half the white and black levels for instance (any case with the same ratio). In this case same shape meaning the same ratios between every level.

Is that not the case?

Did you end up plugging 0.1 in somewhere when running the analysis? If so I would have expected it to need a white level too.

Thanks,
Darin
You are correct, a constant contrast ratio yields identical curves (previous post amended). The best fit Bt.1886 curve to sRGB has a contrast ratio of 1176.
zoyd is online now  
post #169 of 177 Unread 09-28-2015, 05:20 AM
AVS Addicted Member
 
Stereodude's Avatar
 
Join Date: Jan 2002
Location: Detroit Metro Area
Posts: 12,856
Mentioned: 23 Post(s)
Tagged: 0 Thread(s)
Quoted: 2347 Post(s)
Liked: 1507
Quote:
Originally Posted by darinp2 View Post
I'm guessing that 16 bit would calculating higher for a scene like this, but I'm not sure how much higher.
A little bit higher I guess. 2.2 gamma
Code:
Luma/Luminance statistics for NBNW Sample 09/28/2015 07:19:01
Color space: 1 Color range: 1 Reported FPS: 23.976025
Frame , Current time, Fractional minutes, Hour, Minute, Seconds, Ystim[%], U, V, Ystim accum.[%], U accum., V accum., MinLuminance , MaxLuminance , AverageLuminance, MedianLuminance, Average Luminance[%]
0 , 07:19:01 , 0.000000 , 0 , 0 , 0.000000 , 3.635095 , 128.602722 , 126.702744 , 3.635095 , 128.602722 , 126.702744 , 16 , 210 , 16.729876 , 17 , 0.333277
1 , 07:19:05 , 0.000683 , 0 , 0 , 0.041000 , 3.635095 , 128.602722 , 126.702744 , 3.635095 , 128.602722 , 126.702744 , 16 , 210 , 16.729876 , 17 , 0.333277
2 , 07:19:05 , 0.001383 , 0 , 0 , 0.083000 , 3.659961 , 128.618362 , 126.714386 , 3.643383 , 128.607941 , 126.706627 , 16 , 220 , 16.739157 , 17 , 0.337514
3 , 07:19:06 , 0.002083 , 0 , 0 , 0.125000 , 3.650412 , 128.623245 , 126.738380 , 3.645141 , 128.611771 , 126.714561 , 16 , 236 , 16.737000 , 17 , 0.336529
4 , 07:19:06 , 0.002767 , 0 , 0 , 0.166000 , 3.635095 , 128.602722 , 126.702744 , 3.643131 , 128.609955 , 126.712204 , 16 , 210 , 16.729876 , 17 , 0.333277
5 , 07:19:06 , 0.003467 , 0 , 0 , 0.208000 , 3.642215 , 128.629364 , 126.730972 , 3.642978 , 128.613190 , 126.715332 , 16 , 219 , 16.733877 , 17 , 0.335104
6 , 07:19:06 , 0.004167 , 0 , 0 , 0.250000 , 3.659961 , 128.618362 , 126.714386 , 3.645405 , 128.613922 , 126.715195 , 16 , 220 , 16.739157 , 17 , 0.337514
7 , 07:19:06 , 0.004850 , 0 , 0 , 0.291000 , 3.652993 , 128.615753 , 126.743446 , 3.646353 , 128.614151 , 126.718727 , 16 , 219 , 16.736681 , 17 , 0.336384
8 , 07:19:06 , 0.005550 , 0 , 0 , 0.333000 , 3.639846 , 128.625443 , 126.738342 , 3.645630 , 128.615417 , 126.720909 , 16 , 217 , 16.733284 , 17 , 0.334833
9 , 07:19:06 , 0.006250 , 0 , 0 , 0.375000 , 3.639859 , 128.621445 , 126.751106 , 3.645053 , 128.616013 , 126.723923 , 16 , 183 , 16.726986 , 17 , 0.331957
10 , 07:19:06 , 0.006950 , 0 , 0 , 0.417000 , 3.627660 , 128.617630 , 126.745041 , 3.643472 , 128.616165 , 126.725845 , 16 , 89 , 16.724394 , 17 , 0.330773
11 , 07:19:06 , 0.007633 , 0 , 0 , 0.458000 , 3.623763 , 128.644577 , 126.750282 , 3.641829 , 128.618530 , 126.727882 , 16 , 210 , 16.725502 , 16 , 0.331279
12 , 07:19:06 , 0.008333 , 0 , 0 , 0.500000 , 3.648797 , 128.626068 , 126.744804 , 3.642365 , 128.619110 , 126.729179 , 16 , 219 , 16.735428 , 17 , 0.335812
13 , 07:19:06 , 0.009033 , 0 , 0 , 0.542000 , 3.651760 , 128.627991 , 126.742043 , 3.643037 , 128.619751 , 126.730103 , 16 , 220 , 16.737137 , 17 , 0.336592
14 , 07:19:06 , 0.009717 , 0 , 0 , 0.583000 , 3.640723 , 128.628250 , 126.738945 , 3.642883 , 128.620316 , 126.730690 , 16 , 221 , 16.735262 , 17 , 0.335736
15 , 07:19:07 , 0.010417 , 0 , 0 , 0.625000 , 3.650646 , 128.635986 , 126.740631 , 3.643368 , 128.621292 , 126.731316 , 16 , 221 , 16.737526 , 17 , 0.336770
16 , 07:19:07 , 0.011117 , 0 , 0 , 0.667000 , 3.634263 , 128.637939 , 126.739883 , 3.642832 , 128.622269 , 126.731819 , 16 , 221 , 16.734442 , 17 , 0.335362
17 , 07:19:07 , 0.011817 , 0 , 0 , 0.709000 , 3.647061 , 128.636734 , 126.747726 , 3.643067 , 128.623077 , 126.732704 , 16 , 221 , 16.736994 , 17 , 0.336527
18 , 07:19:07 , 0.012500 , 0 , 0 , 0.750000 , 3.630250 , 128.634949 , 126.743866 , 3.642392 , 128.623703 , 126.733292 , 16 , 219 , 16.733248 , 17 , 0.334816
19 , 07:19:07 , 0.013200 , 0 , 0 , 0.792000 , 3.633796 , 128.645645 , 126.755379 , 3.641963 , 128.624802 , 126.734390 , 16 , 218 , 16.733280 , 17 , 0.334831
20 , 07:19:07 , 0.013900 , 0 , 0 , 0.834000 , 3.641787 , 128.629807 , 126.739273 , 3.641954 , 128.625031 , 126.734627 , 16 , 220 , 16.735138 , 17 , 0.335679
21 , 07:19:07 , 0.014583 , 0 , 0 , 0.875000 , 3.630100 , 128.642578 , 126.754654 , 3.641415 , 128.625839 , 126.735535 , 16 , 222 , 16.732361 , 17 , 0.334411
22 , 07:19:07 , 0.015283 , 0 , 0 , 0.917000 , 3.640682 , 128.626724 , 126.739822 , 3.641384 , 128.625870 , 126.735725 , 16 , 223 , 16.734699 , 17 , 0.335479
23 , 07:19:07 , 0.015983 , 0 , 0 , 0.959000 , 3.650465 , 128.634720 , 126.743156 , 3.641762 , 128.626236 , 126.736031 , 16 , 221 , 16.736811 , 17 , 0.336443
End
Stereodude is offline  
post #170 of 177 Unread 09-28-2015, 05:31 AM
AVS Addicted Member
 
Stereodude's Avatar
 
Join Date: Jan 2002
Location: Detroit Metro Area
Posts: 12,856
Mentioned: 23 Post(s)
Tagged: 0 Thread(s)
Quoted: 2347 Post(s)
Liked: 1507
Quote:
Originally Posted by darinp2 View Post
Do you happen to have Guardians in the Galaxy? I used the left eye from the 3D version for that one, but not sure how close it is to the 2D version.
I re-ran Guardians of the Galaxy 2D for 2.2 gamma. This time I sampled every frame, not just 1 frame per second.

Guardians of the Galaxy (2D) [all frames / 2.2 gamma]:
16 bit mean, median, max: 5.762%, 2.444%, 91.407%

Black bars were cropped since the 2D release is a constant 2.40:1, credits were kept.
Stereodude is offline  
post #171 of 177 Unread 09-28-2015, 12:09 PM
AVS Addicted Member
 
darinp2's Avatar
 
Join Date: Sep 2003
Location: Seattle, WA
Posts: 23,085
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 890 Post(s)
Liked: 1008
Quote:
Originally Posted by Stereodude View Post
A little bit higher I guess. 2.2 gamma
Thanks. So, looks like with this dark scene I got about 1/600th of white with the 8 bit version and you got about 1/300th of white with the 16 bit dithered version.

--Darin
Ericglo likes this.

This is the AV Science Forum. Please don't be gullible and please do remember the saying, "Fool me once, shame on you. Fool me twice, shame on me."
darinp2 is offline  
post #172 of 177 Unread 09-28-2015, 04:48 PM
AVS Addicted Member
 
Stereodude's Avatar
 
Join Date: Jan 2002
Location: Detroit Metro Area
Posts: 12,856
Mentioned: 23 Post(s)
Tagged: 0 Thread(s)
Quoted: 2347 Post(s)
Liked: 1507
Quote:
Originally Posted by Stereodude View Post
I re-ran Guardians of the Galaxy 2D for 2.2 gamma. This time I sampled every frame, not just 1 frame per second.

Guardians of the Galaxy (2D) [all frames / 2.2 gamma]:
16 bit mean, median, max: 5.762%, 2.444%, 91.407%

Black bars were cropped since the 2D release is a constant 2.40:1, credits were kept.
I did the same for the "3D only" disc.

Guardians of the Galaxy (3D / Left Eye) [all frames / 2.2 gamma]:
16 bit mean, median, max: 7.793%, 3.595%, 85.375%

140 pixels were cropped top and bottom for consistency with the 2D release, credits were kept.
Stereodude is offline  
post #173 of 177 Unread 09-29-2015, 07:00 PM
AVS Addicted Member
 
darinp2's Avatar
 
Join Date: Sep 2003
Location: Seattle, WA
Posts: 23,085
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 890 Post(s)
Liked: 1008
I touched on a little of this in the shootout thread here, but I've been thinking about how HDR will change the trade-offs between good on/off CR and good ANSI CR.

Looking at the values we've calculated with SDR content we can see that in general the median images generally have much lower ADLs than the 50% of the ANSI checkerboards. The relationship between how much on/off CR matters compared to ANSI CR is rather complicated, but the makeup of the images viewed has some bearing on this, with low ADL images tending to lean more toward on/off CR mattering more than with many of the higher ADL images where good ANSI CR can show more benefit.

If we get to analyze HDR content as encoded I think we will find that compared to SDR the median values will be lower when calculated as a percentage of the highest white point, which I believe is what matter most here since these projectors don't have zones and their black floors at any point in time are limited by their peak white divided by their native on/off CR at that point (such as that iris position). The reason I expect the median values to be lower is because I expect HDR will tend to be encoded to make the highlights stand out and that requires separating the highlights from the average parts more.

Essentially, for projectors I think HDR content will tend to skew the trade-offs between good on/off CR and good ANSI CR more toward on/off CR than with SDR content.

--Darin
Ericglo likes this.

This is the AV Science Forum. Please don't be gullible and please do remember the saying, "Fool me once, shame on you. Fool me twice, shame on me."

Last edited by darinp2; 09-29-2015 at 07:04 PM.
darinp2 is offline  
post #174 of 177 Unread 10-01-2015, 01:32 AM
AVS Addicted Member
 
darinp2's Avatar
 
Join Date: Sep 2003
Location: Seattle, WA
Posts: 23,085
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 890 Post(s)
Liked: 1008
Thanks Soulnight and good job. I really appreciate it.

You mentioned the EBU test pattern. One of the issues I have with that one for taking measurements off a screen with a front projector is trying to figure out how to do that accurately. For instance, just measure the black part without the gray part around it.

For measuring from the projector I don't think that is much of an issue because a meter can be placed just in the right spot, but a meter measuring off a screen can be much more difficult without getting in the image and causing reflections, etc.

For the test patterns you mentioned do you envision those being used while measuring off the screen? If so, do you think the meters we can reasonably get can measure the white part accurately? I tried a multi-thousand dollar meter and am not confident it can do it. If measurements are including just a little bit outside the white area then the white values are being reported too low. One test could be to measure the white on a full white screen and one of these test patterns from the projector, then do the same off the screen and see if the ratios are fairly consistent.

For my recommendation with my presentation to SID I decided to go with the larger area that is in the corner box contrast test which is already a standard in the IDMS. That one is 20% of the width and height, for 4% ADL. I lowered the video level of that to 20% in order to get a low ADL test pattern for two reasons. One is my concern about measuring small areas off screens and the other is because dynamic irises can provide some benefit to lower ADL images that don't contain any pixel encodings near 100% video level for any of the primaries.

Your test patterns seem to largely go after getting something approaching native on/off CR with your lowest ADL pattern and that is important to know.

The dynamic iris things get complicated since there are images they can't improve the intra-image CR for and images where they can. As a simple example, if a projector had 2k:1 native on/off CR and reasonable ANSI CR then it should get close to 2k:1 with your 1% ADL patterns. If a dynamic iris is turned on that just lowers the white and black level by 50% the projector would still only get close to 2k:1 with your 1% ADL patterns, which is far from what a native 4k:1 native on/off CR and reasonable ANSI CR projector could do.

However, there are images where the 2k:1 native on/off CR with 2x dynamic iris can do as well as the 4k:1 native on/off CR projector with no dynamic iris. For instance, if the brightest pixel for any primary is 50% video level (~20% luminance) then that dynamic iris projector can just close the iris 50%, apply a 2x adjustment to the gamma and have everything at the same luminance level (including black) at the screen as the 4k:1 native on/off CR projector. There is more to this (like if the DI projector is a DLP it has more time resolution for moving those mirrors to the right level near black, where DLPs struggle and generally have to dither) and there are definitely trade-offs and limits with DIs.

It might make sense to think of this like eShift, where when done right a 2K projector with eShift has cases where it is more like native 2K and cases where it is more like native 4K, just depending on which image is chosen.

The data you've gathered shows well that ADL for those movies are far from the 50% ADL of the ANSI checkerboards and I think provides good support for why native on/off CR matters. Just from looking at how much stuff is pushed toward the zero ADL I say we have a good idea that some improvement could be made to the intra-image CRs for some of those images through use of a dynamic iris. Some of those images are the type with no bright pixels where a DI could do very well and some of them are the type with some bright pixels that limit what a DI can do before causing more harm than good.

When Mark Petersen came up with some test patterns for his threads that proceeded this one I believe he used some patterns that had white on black (where DIs would struggle) and some test patterns that modified ADLs by modifying the highest video level that went with black (where DIs can do more).

Thanks,
Darin
Ericglo likes this.

This is the AV Science Forum. Please don't be gullible and please do remember the saying, "Fool me once, shame on you. Fool me twice, shame on me."
darinp2 is offline  
post #175 of 177 Unread 10-02-2015, 08:53 PM
AVS Addicted Member
 
darinp2's Avatar
 
Join Date: Sep 2003
Location: Seattle, WA
Posts: 23,085
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 890 Post(s)
Liked: 1008
I was thinking about how often a dynamic iris could help the CR with some of these movies and what information we would need to gather. Things can get pretty complicated with images with pixels near white intermixed with images with no bright pixels (the DI algorithms need to avoid visible pumping) and it would be good to figure out what the brightest level for any primary is, not just the brightest average luminance for all 3 primaries combined.

That said, I figured it might be interesting to look at a little bit of the data I had already gathered in an excel spreadsheet. I just added one entry for the median of the peak luminances for all frames.

I just checked North By Northwest not counting the credits and got 79% for the median of the peak luminances.

I then entered starting points and ending points for the Mount Rushmore scene that takes up about 6 minutes at the end and got 41%. This is for frame 186246 to frame 195500. This was a section where I got a median ADL of 1.3%.

Ignoring the complications I mentioned above and just taking the 41% as the brightest pixel for any primary for this thought experiment (even though it isn't a valid assumption), if a dynamic iris was to shut down 50% for all the frames with 41% peaks and below the video could be remapped up to 82% with a black floor of half of what it would be without the dynamic iris (not even counting that DIs can increase native on/off CR as they close). For that 6 minute segment this doubling of effective on/off CR I would say that half the frames could get improved intra-image CRs with little to no negative visible side effects.

This is one reason that the position of some in this industry that we only need to look at ANSI CR is so wrong. Turning the DI off or on would make no difference to the ANSI CR, but would make nice improvements to the intra-image CRs of many of these real images. Of course, some manufacturers see an opportunity and end up pushing things far enough to cause more harm than good, but the opportunity is definitely there for more good than bad when done right, as I think many people would agree to based on real world experience watching with their own eyes (not just relying on a meter to tell them whether something will look good or not).

The above is also one reason that it isn't just the ratio between white and black at the same time that matters, although that is important.

I see the fact that many people will prefer some artifacts with a DI over turning the DI off and living with the higher black floor as a testament to how much the black floor matters.

Mining the data for when a DI can improve things visibly and when they are likely to run into problems could likely get pretty complicated.

--Darin
Ericglo likes this.

This is the AV Science Forum. Please don't be gullible and please do remember the saying, "Fool me once, shame on you. Fool me twice, shame on me."
darinp2 is offline  
post #176 of 177 Unread 11-01-2015, 03:25 AM
Senior Member
 
Soulnight's Avatar
 
Join Date: Dec 2011
Posts: 210
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 107 Post(s)
Liked: 77
Thanks Darin for your comment on our ADL study!

Your thoughts on the use of dynamic iris depending of the maximal brightness found in on picture are very interesting.

If we find some time, we will look into that as well.

Keep the good work!

Anna&Flo
Soulnight is offline  
post #177 of 177 Unread 11-17-2015, 11:37 AM - Thread Starter
AVS Addicted Member
 
stanger89's Avatar
 
Join Date: Nov 2002
Location: Marion, IA
Posts: 19,376
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 1287 Post(s)
Liked: 579
Hey Soulnight, I didn't see it anywhere, do you have those patterns available anywhere for download?

See what an anamorphoscopic lens can do, see movies the way they were meant to be seen
stanger89 is offline  
Sponsored Links
Advertisement
 
Reply Digital Hi-End Projectors - $3,000+ USD MSRP



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