Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

CHARTS! CHARTS! CHARTS! too many Charts, Too confusing,...

OptiTek said:
OK, One more time... They stated in the article that the 76 (5%) lines are aliasing.but that is NOT what I'm asking. What I wonder is how they got more than 1080 lines out of that camera. I thought the output was in camera super sampled down to 1080 so you never get more than 1080(2K) or so...
Click to expand...

It's not that 76 lines are aliasing, but that "at least 76 lines are aliasing" and given our knowledge of lenses, sensors and OLPFs, the "at least" is really "probably a lot more than 76 lines are aliasing". The Alexa has a raw mode that records 2880 pixels across. Typical OLPF and demosaic leads to around 75% with typical algorithms about 80% with good ones (what Deanan above referred to as the Bayer Efficiency) so any figure above 2304 is suspect, and depending on other factors, the actual figure could be a lot lower.

Graeme
 
Graeme Nattress said:
It's not that 76 lines are aliasing, but that "at least 76 lines are aliasing" and given our knowledge of lenses, sensors and OLPFs, the "at least" is really "probably a lot more than 76 lines are aliasing". The Alexa has a raw mode that records 2880 pixels across. Typical OLPF and demosaic leads to around 75% with typical algorithms about 80% with good ones (what Deanan above referred to as the Bayer Efficiency) so any figure above 2304 is suspect, and depending on other factors, the actual figure could be a lot lower.

Graeme
Click to expand...
Thanks Graeme, so the tests you posted in the other thread had another RAW mode that outputted 1080(2K) signal?
Is there more than 1 raw mode in the camera? I realize I'm asking you about a competitor but there is a big discrepancy between the two tests and while your findings made sense for the 2K camera this 3K result threw a monkey wrench into things....
I'm trying to wrap my head around this as I get this questioned a lot in relation to Optimator, lenses, etc... I would love to have a clear understanding of this....
 
OptiTek said:
Thanks Graeme, so the tests you posted in the other thread had another RAW mode that outputted 1080(2K) signal?
Is there more than 1 raw mode in the camera? I realize I'm asking you about a competitor but there is a big discrepancy between the two tests and while your findings made sense for the 2K camera this 3K result threw a monkey wrench into things....
I'm trying to wrap my head around this as I get this questioned a lot in relation to Optimator, lenses, etc... I would love to have a clear understanding of this....
Click to expand...

The discrepancy is what makes the SCCE results confusing.
For the leica test we shot a zoneplate which shows aliasing very clearly.
If you use something like a trumpet, the results can easily be misleading
as it's harder to see resolution vs aliasing.
 
OptiTek said:
Thanks Graeme, so the tests you posted in the other thread had another RAW mode that outputted 1080(2K) signal?
Is there more than 1 raw mode in the camera? I realize I'm asking you about a competitor but there is a big discrepancy between the two tests and while your findings made sense for the 2K camera this 3K result threw a monkey wrench into things....
I'm trying to wrap my head around this as I get this questioned a lot in relation to Optimator, lenses, etc... I would love to have a clear understanding of this....
Click to expand...

The tests we posted were from 1080p ProRes and 2880x1620 RAW. Why the discrepancy between a resolution figure that cannot be right because it is greater than 2880, and a full MTF plot on a sine zone plate that shows resolution and, more importantly, aliasing, quite clearly? Deanan's given the answer - when you confuse aliasing for resolution you can get incorrect results. Graeme's answer - because we measure resolution properly, it being something we specialize in.

Has anyone actually said what kind of rez charts they used? Did they check those charts were accurate or just assume? Too many questions, and the only answers we have are demonstrably wrong.

Graeme
 
Thanks for your replies.
Let me put it this way:
1 epic M- $58K
1 epic x-28K
Getting the two main tech brains to answer my forum Questions- PRICELESS!!!!!
 
Of course, you get more than just the camera for your $58K on the Epic-M though, but I do get your point :-)

Graeme
 
Charting a course

Charting a course

As a data point, the zone plates used in the SCCE were actually 4 different concentric circles with different colors chosen for each set of pairs.

It makes sense that a black/white pair would more accurately reveal the performance of the luminance information as derived from the green filtered photosites plus the interpreted luma values of the red and blue filtered ones. As a cameraman I want to understand the monochromatic resolution of the system, but if the juxtaposition of specific colors creates unusual moire response that could be important as well. What is perhaps being shown in greater granularity when using 4 different color pairs is the limitations of a particular debayering algorithm as a variable in achievable resolution when using a Bayer pattern on the sensor.

I find it interesting that as people like Graeme and the color scientists at Arri, Sony, et al dig deeper into the strengths and weaknesses of the Bayer CFA architecture there is a concurrent need to develop meaningful ways to measure the results. Many of us jumped on Sony for claiming efficiencies from the Q67 (rotated Bayer) tech that defied orthodoxy, but time will tell if advances in decoding will show a greater performance envelope for either geometry. That said, Sony's marketing driven presentation of the tech was disingenuous at the least and obfuscation if you ask me - but I digress.

While it is easy to just say "let the images speak for themselves" I believe that there is much to gain from understanding the technology on a deeper level in seeking specific image metrics to reach creative goals. At the risk of sounding like a suck-up, Graeme's tagline about the relationship between stories and science is spot on in my book.

I think its important to remember that what the sensor "sees" is not anything like what the human eye sees and turning that data set into an accurate representation of what "normal" humans see requires a number of steps.

Just 2 cents from a chart lover ;-), hope to see many of you tomorrow at RSH-

Cheers - #19
 
Blair, do you know anything more about the charts they used? I often used different coloured zones for checking demosaic algorithms, and it's a worthwhile thing to do, but a monochrome chart shows things up too, if you know how to look at them.

I'm pleased you picked up on my tag line - it's there for a reason.

Graeme
 
From what I understand from DSC, that chart was only used for visual comparison, not measurement. I still need to get details on the erroneous measurements and where they came from.

Graeme
 
Thanks for the explanations, Graeme. I follow your posts almost religiously. I didn't get to see these tests at NAB, and I didn't get to meet you either, I'm afraid to say. Although I'm still learning all this testing, I'm concerned that people who appear that they should know better are getting results that even to me are obviously wrong.

Albert
 
Thanks Albert. I'm sure we'll get to the bottom of these tests....

Graeme
 
Hi Graeme,

thanks for investing that much time into Reducation here at Reduser! This thread and the Leica one are obligatory reading for my students now.
 
Graeme, you might try reaching out to Mike Curtis if you haven't already. No offense to Mike, or Bob Primes, or the rest of the dedicated craftspeople who worked on the SCCE, but, like all tests, there were choices made in methodology that influenced the results.
Monday morning quarterbacking is easy, sacrificing huge blocks of time is hard, that must be said. Still, deconstructing the results is a valuable step in moving the knowledge base forward.

Cheers - #19
 
I've spoke to Mike, and I see that on PVC the tester has posted too:

"The Maximum Resolution numbers were determined using the slanted edge method in ISO 12233." - ok, now we know the method.

"The lp/ph numbers are the average of horizontal and vertical measurements (although these were very close for all cameras measured). The method allows spatial frequency response above Nyquist to be measured, although such frequencies will alias." - this is where we come un-stuck:

The H and V resolutions for F35 and the DSLRs are not remotely similar. We've all seen that and it's especially clear on the DLSRs. If "such frequencies will alias" then they are not resolution and should not be counted as such. To promote publicly such figures as maximal resolution figures beyond that even the sensor is capable of delivering is just not right. Of course, it's not so much that such frequencies will alias, but that such contrast on an MTF plot is caused by aliasing. I could almost understand if they displayed a full MTF plot and marked on where visible aliasing sets in, but even the chosen method of calculating MTF is one which is designed to avoid aliasing issues.

"Also, the Alexa and PhantomFlex numbers in the initial report are preliminary values that will be updated." - so preliminary numbers went out to the public, and those numbers are wrong.

There's a reason we measure resolution the way we do:

1) it's a direct measurement from the chart - you don't need software to interpret the data
2) You can easily see aliasing
3) You can easily see and any difference in horizontal and vertical resolution
4) Because you can see the aliasing, you'll never give a max resolution figure that's silly.

Graeme
 
It does seem like they are counting a lot of false detail / aliasing as valid resolution. There are major holes beyond the resolution chart too. These charts are clearly not ready for public perusal and it is irresponsible of the SCCE (or whoever concerned) to publish them.
 
Subhadip, there probably are similar holes in the other tests they performed. Resolution is one of the easiest things to measure on a camera and one of the hardest to screw up, so when you see problems there, it does get you concerned. From what I understand thousands of camera people saw these tests at NAB, and what will be the numbers they take home with them? They'll take these numbers that are wrong, and that's bad for the general camera person, bad for the industry and bad for the organizers of the tests when they have to explain why they were presenting resolution figures that are wrong.

Graeme
 
Graeme,

I also believe there is an issue w/ how resolution charts are printed. IIUC, the transition between black and white lines should be gradual and sinusoidal. In contrast (no pun intended) some rez charts transition immediately from black to white w/ no gradation.

I read that one of the charts they used was a colored one called "The Wringer." I believe The Wringer is not sinusoidal and hence shows higher resolution than would a sinusoidal chart.
 
I know of the wringer - it's a chart for visualizing details at different colours, not a chart for measuring resolution on, and it was, I'm told, only used for visual comparisons. We use sine charts indeed for the reasons you specify. However, the method they used is the slanted line approach which needs a hard transition to work, and then the spread function of that transition is measured and from that a MTF is computed. It's not a direct measurement technique but a calculated technique, and it's also designed to measure MTF beyond sampling theorem limits, so by it's very nature it will confuse aliasing with resolution and give you a limiting resolution figure that's bogus. As pointed out above there's some very strong reasons why we test resolution the way we do - we've found them simple, repeatable, reliable and they're a direct rather than a computed measurement, which makes it really easy to spot if you've made a mistake.

Graeme
 
You must log in or register to reply here.
Back
Top