New Blackmagic Camera July 16th?

[Joshua Cadmium]

Interesting analysis, Joshua. Perhaps the implication is that JB might typically use a fatter stop than the moire sufferers?

Relying on diffraction to suppress moire would seem to defeat the purpose of having such a high-resolution sensor. Though perhaps an OLPF does as well.

At any rate, the matter is probably best settled as most lens/camera issues are - by testing and seeing if you're happy with the outcome and the gestalt.

Yes, testing is essential especially because this sensor is not Bayer, so we can't really compare this sensor to anything else when it comes to moire. It has more red and blue pixels than an equivalent Bayer sensor, but if the color sampling is not as evenly spaced as it is on a Bayer sensor, color moire might even be worse. But if John's not seeing it that much, it shouldn't be as big of a deal.

I brought up using diffraction to suppress moire for two reasons. One, it seemed like a potential solution if it ever showed up. Because the pixels are so small, they are going to blur together more than on any other cinema camera.

The second reason I posted so much info, was that it has implications in terms of capturing all of the resolution of the 12K. If you want to punch in to 4K, or even HD, you need to realize that diffraction is going to cause those punch ins to be less sharp than they could be if you are at anything slower than maybe f2 (but people need to test).

If you are just going to be downsampling 12K to 4K or 8K, it's not going to matter as much, but it is something to be aware of.

 

[Joshua Cadmium]

Do you think that since the 12K sensor can have a lower chroma subsample, that may actually moire more than a similarly spec’d bayer (pronounced ‘buyer’) colour array? Not saying that it isn’t easily mitigated though, and the extra pixel density will always have some unique properties. But adding enough dithering to prevent moire, would lower the light sensitivity and resolution to the point that there is little advantage over an 8K bayer sensor?

We don't know the pattern of the sensor, so I don't think there is a way to definitely answer that - as I said in the previous post it could moire more or less, but maybe less.

Was the pixel density an important part of the 12K design, or could it be an advantage to have gone LF with the sensor?

The small pixel size and the CFA pattern go hand in hand, as smaller photosites have a smaller bucket to hold on to photons they capture and can lead to less dynamic range, which is why they are using white photosites to mitigate that. Arri is still using 8.25um photosites because they are huge buckets to hold on to lots of photons. Blackmagic could have gone with bigger pixels at a 12K LF size, which would have been better from a dynamic range standpoint, but then you wouldn't get the amount of lens options you get with this sensor - there are a lot of choices in S35. But, they could also go another direction (with I bet they'll do) and do a LF size in 16K. It makes more sense from a business standpoint since they already developed the sensor tech at 2.2um.

I don’t mind if an OLPF reduces resolution, as the extra chroma and luma sampling will still take place, making for some lovely tones and gradations. However if an olpf dithers enough light, does that effect the senstitivity? Or is that a non-issue.

If by sensitivity you mean dynamic range, it wouldn't - it's just spreading the point source of light out a bit, but it's doing so everywhere. Yes, even though resolution takes a hit, the extra sampling still takes place.

 

[John Brawley]

Josh.

Like a circle of confusion isn’t absolute, when making decisions about depth of field calculations, there are generally accepted numbers that are de facto standard.

2.5x is the defacto number for the Airy disk calculations.

Also, BMD lay out the array on their patent. Also think about the difference between pixels and photo sites. And how these are changed and re-arranged per the BMD patent with BMD raw.

https://patentimages.storage.googleapis.com/88/08/39/428dbbc9e5dfca/US20190306472A1.pdf

I typically shoot around T2-T4.

Most of my demo material that was released that you all can download is shot around T2.8.

JB

 

[Guest]

If by sensitivity you mean dynamic range, it wouldn't - it's just spreading the point source of light out a bit, but it's doing so everywhere. Yes, even though resolution takes a hit, the extra sampling still takes place.

I meant, as in sensitivity, like asa or iso rating. Wondering how much diffusion slows down the light.

 

[Joshua Cadmium]

It’s a contender for sure, as an A camera option. The main things holding it back are those high end utility things. Not anything to do with IQ.

That's great to hear.

The mini of course has internal switchable NDs and that’s what I’m more referring to.

Gotcha. I forgot about that. I was thinking of the internal filter module for the older Alexas.

This idea that IR is simple is just...wrong.

Like with audio EQ. You can’t really just totally cut below a certain frequency. You can HEAR the difference. Same with IR. You have to roll it off to some degree.

In our early days of ND they were super aggressive. Everyone accepted very GREEN ND filters because there was no IR and that was the price you paid.

What I meant was not IR, per se, which does have to be rolled off at the sensor level and is not a simple adjustment. I was actually referring to completely neutral NDs. It SEEMS like it would be an easy thing to just truncate every nanometer evenly across the board - just stop a certain amount of light evenly across the whole visible spectrum, but it is not easy. Of course thinking it is easy is wrong - that's not what I was saying if that's what you were thinking. What I was trying to communicate was that I find it ironic that one of the simplest things you have to do to control the image, and the filters you always have to have on hand - ND - has to be such a complicated affair. For photography, it is such a simple thing to just change the shutter speed, but our hands are mostly tied regarding that.

Oh, yeah those green Tiffen IRNDs didn't look great. You could balance them back to white, just like the bluish Schneider Platinum NDs could be balanced back to white, but it was for sure a compromise. Of course you had and still have Pancro NDs, which have been around longer than any digital camera Arri or Red have put out. They were, and maybe still are the best, but they are a reflecting mirror and don't get into deep ND - you have to custom order anything beyond a 1.5. On this video that Nisi put out a few years ago, I thought the Pancro may have looked the best, especially with tungsten light: https://www.youtube.com/watch?v=3Hpx9X-ePow .

The idea of vintage lenses and diffusion mitigating moire is a bit of a red herring. An OLPF is designed to reject detail that is too fine for the sensor to accurately reproduce. Diffusion filters are the opposite. The soften the detail that’s within what the sensor CAN resolve. Same with vintage lenses.

I do have to disagree with this as well. You can have a lens that is so poor in resolution that it will limit aliasing. There is a mathmatical example here, with a Siemans Star comparison at the end of the article: https://www.lensrentals.com/blog/2017/10/the-8k-conundrum-when-bad-lenses-mount-good-sensors/ .

The same thing is true with some, but definitely not all, diffusion filters. Anything that blurs real resolution is also going to blur false resolution, otherwise an OLPF wouldn't work in the first place. To what extent you get that blurring would need to be tested, but I would bet something like a Tiffen Double Fog or a Mitchell Diffusion filter would cut down on some level of moire, almost certainly not all, but some. A Pearlescent or Radiant Soft filter? Maybe not so much at all.

The only thing I can think of that really affects this in camera is the shooting stop (and thus dof).

If you mean shooting wide open, then yes, there can be less resolution there, but some lenses are effectively diffraction limited and will still give enough resolution to show moire at lower f stops. And, the more the photosite size shrinks, the more this is going to happen.

That’s exactly what’s great about the machine I listed. It’s very consistent. And that leaves the variables to bean...roast...batch etc....

The complexity of coffee is lovely, isn't it? Species, terrior, and the Maillard reaction produce such stark differences in flavor.

 

[Guest]

What I was trying to communicate was that I find it ironic that one of the simplest things you have to do to control the image, and the filters you always have to have on hand - ND - has to be such a complicated affair.

+1

It would seem that for professional cameras, adding a couple grand for the option of having the best filtration internal would be ideal.

 

[rizibo]

12K URSA Mini Pro VS 8K Canon R5 VS 6K C500 MKII:

https://www.youtube.com/watch?v=3JGvfbJXZkk

What is causing the yellow cast on the model’s hair filmed on the Ursa at 6:36? The canon cameras don’t have this.

 

[Joshua Cadmium]

Like a circle of confusion isn’t absolute, when making decisions about depth of field calculations, there are generally accepted numbers that are de facto standard.

2.5x is the defacto number for the Airy disk calculations.

As I mentioned, one can still use 2.5x photosite widths, just like you could use d/1500 as a circle of confusion number. However, knowing where that 2.5x number comes from is important, because diffraction will start before that 2.5x number, as Airy Disks are starting to blur together as soon as they cross the individual photosite threshold.


Even using that 2.5x number, it puts the 12K sensor into diffraction at a width of 2.5 * 2.2um or 5.5um. That means that the 12K sensor will be in visible diffraction at f4.1 (2.44 * 550nm * f4.1 = 5.5um.) That means at about f4, you are definitely going to see diffraction, especially if you punch in. That is still much less than I think anyone would realize.


But that 2.5x ratio is misleading.


Why? Let's look at some real numbers again and for that I'll use that Zeiss Batis 85mm again, since it best proves my point: https://www.opticallimits.com/sonyalphaff/1032-zeissbatis85f18?start=1 . On the A7RII using 2.5x photosite widths, that defacto number is saying that diffraction will only be experienced at an Airy Disk width of 2.5 * 4.5um = 11.25um. That means that the A7RII will be in visible diffraction at f8.4 (2.44 * 550nm * f8.4 = 11.25um.)


This online calculator, which is the first result when I search for "diffraction calculator" tells me that the A7RII at 42.4MP is diffraction limited at f9 (they rounded up): https://www.photopills.com/calculators/diffraction . If I choose f8, it specifically says that I am not diffraction limited.


So, you think, well, this defacto number is telling me that I will be in visible diffraction at f8.4 (according to the math, or f9 according to that online calculator.) I'll just stop down to f8 - that's less than f8.4.


What happens to resolution at f8 on that Zeiss Batis? The resolution in the center is 4773, which is 5.7% less resolution than the maximum center resolution of 5063 at f2.8. At the near center, there is 6.0% less resolution. At the border there is 7.8% less resolution and at the edges there is 6.3% less resolution. Doing a rough estimate, let's just say that there is 6% less resolution everywhere. Well that scales in both directions, so total resolution loss would be 1 - (0.94 * 0.94) or 11.64% less resolution everywhere. That's a loss of about 5 megapixels on at 42.4 megapixel sensor.


[Also, we were comparing f8 to f2.8, when the sharpest this lens would have gotten would most likely be where the Airy Disk meets the photosite size, which it would at f3.35 (2.44 * 550nm * f3.35 = 4.5um).]



--------



So, what if you looked at that f8.4 number and said, well, that's awfully close to f8, so I'll just avoid f8 all together and live at a f5.6 limit. This calculator even tells me for full frame 42.4 MP, I won't see diffraction until f6.8 (2.0x width): https://www.cambridgeincolour.com/tutorials/diffraction-photography-2.htm


At f5.6, you would lose a little over 4% of resolution in the center (in one direction - about 8% total) and about 2% in the edges (about 4% total) as compared to the maximum resolution the lens can see at f2.8.


Can you see how the 2.5x and even 2.0x ratios can be misleading? They are not really where diffraction starts. I want to know where I am maximizing resolution and then go from there.


Looking at the 12K, if you took that 2.5x width as gospel and shot at f4, you may see somewhere between 10-15% less resolution than you would otherwise get. That's about 10 megapixels off the top. Resolution isn't everything, but if you wanted to maximize resolution, you may want to know where that actually happens. Or, the other side of the coin - you may want to know where could you stop down and maybe reduce moire because total resolution is dropping.



--------



Another way to illuminate the situation is by looking at a poorer lens that can't resolve as well. Here is the Sony FE 50mm f1.8: https://www.opticallimits.com/sonyalphaff/980-sonyfe50f18?start=1 . Not a great looking resolution chart. This lens has the center peak at f4, since the aberrations being corrected from stopping down does more to boost resolution upwards than the level of diffraction pulling the resolution downwards. It's the same with the edges - they peak at f5.6 because they have such strong aberrations they need to be stopped down a lot in order to correct them - so stopping down draws resolution upwards, more than what diffraction is pushing downwards.


If we went by that 2.5x, it would say we should see diffraction at f8. And we do. But the diffraction started in the central portion earlier than that - at f5.6. If the lens had poorer central resolution, then diffraction would not be limiting it at f5.6. But, since it does have a relatively sharp center, that means that diffraction is kicking in earlier than it does to the poorer edges. And if we compare the center portion between f8 and f4, it's a loss of about 20% of resolution in the center with the edges nearly the same.


So, even if a lens will not resolve the total sensor, you are still going to lose resolution at that 2.5x photosite width (except maybe if you are using really poor resolving lenses.)


This is also why I think the 2.5x used to be relevant - we were dealing with lenses that weren't nearly as sharp. If that Sony 50mm had a weak center then f8 is where diffraction would show and the 2.5x would hold. (But, even if the 2.5x held, that does not say exactly where diffraction kicks in.)



--------



So, to put this all another way - the 2.5x ratio for sure tells you that you are in diffraction - but you have hit that diffraction mark way before you think you did - and the online calculators are all misleading in that regard. It is certainly not the very edge of diffraction, which is what I was trying to determine. Once you are at 2.5x, you are automatically losing info, so if you are trying to preserve resolution, you should stay away from that ratio.


To put this a second way - at what point do you know you are maximizing the total resolution of the system? That point is when the central portion of the Airy Disk equals one photosite width. When you stop down a lens, the Airy Disk gets bigger and spills over into other photosites, blurring the Airy Disk. You might have a little wiggle room, but not much.


So, to sum up, on modern super sharp lenses, you are going to maximize total resolution on the 12K sensor somewhere between f1.64 and f2. With poor lenses, you are going to maximize total resolution somewhere after that, maybe around f2.8. Diffraction kicks in after that. This will all be relatively easy to test.


But, you don't have to. This is all physics and can be seen on thousands of lens tests out there. We're not really breaking new ground here, just interpreting existing info.



--------

Also, BMD lay out the array on their patent.

They have two different CFAs on the patent, so I'm still not sure which one they're actually using.

Also think about the difference between pixels and photo sites. And how these are changed and re-arranged per the BMD patent with BMD raw.

Diffraction doesn't go away just because there is a CFA. The photosite is the physical boundary for the Airy Disk. If you want to capture 12K worth of Airy Disk central points of light, you need 12K worth of photosites. If the Airy Disk central points of light get bigger, you no longer can capture as many of them. (And if the Airy Disk central points of light aberrate from a perfect point source to a something not a perfect point [AKA what all lens aberrations do] you also can no longer capture as many of them.) Constructing pixels from photosites isn't going to increase the physical boundary of the photosite.

Now, you might be talking about color resolution, which is a different tangent, but my argument has been about total resolution. Color is subsampled and is going to have a lower resolution, but that's where the pattern of the CFA plays a role, which, again, I'm not completely sure about. I still don't even know if they are using the white pixel to interpolate color, and that's going to play a big role in determining overall color resolution.

 

[Guest]

...Now, you might be talking about color resolution, which is a different tangent, but my argument has been about total resolution. Color is subsampled and is going to have a lower resolution, but that's where the pattern of the CFA plays a role, which, again, I'm not completely sure about. I still don't even know if they are using the white pixel to interpolate color, and that's going to play a big role in determining overall color resolution.

It was the second layout, i believe. They state it is similar to the Bayer array pattern, but for everyone colour on Bayer, they have two luma(white) and two colour patches.

Bayer type sensor might have one Red photosite, but the BM12K replaces it with:
R=
r w
w r

That is only significant to get the extra light sensitivity and resolution, but the actual chroma subsample is a out quarter resolution, unless they are interpolating values by adding and subtracting values to figure out “virtual” chroma samples.

Bayer would be:
R G
G B

And BM12k is (as seen in figure 4 in the patent doc):
r - w - g - w
w - r - w - g
g - w - b - w
w - b - w - g


Those colour groupings mean that without really smart interpolation, not sure how the base chroma sampling is much more than a 6K bayer sensor. Obviously there is additional tonal gradation, but as far as the groupings and cross talk between photo-sites go, and the potential for moire would be similar to a 6K sensor (quarter resolution=half K) that has an olpf made for a 12K sensor.

But that is misleading, as there are ways to interpolate the chroma values for additional resolution, though interpolation has its own anomalies or side effects, correct? but i won’t pretend that i can synthesize that algorithm in my mind.

Forgive the over simplification, as i am also trying to wrap my head around the pros and cons of this type of sensor design. My personal preference would be for more DR typically, but not every camera has to have the same design. I’m quite liking the bm12k overall, just trying to filter (pun intended) all the hyperbole.

 

[John Brawley]

Josh, I appreciated the detail and depths you're going to, but remember you've already contradicted yourself. Remember when I said ALL cinema cameras can moire even with an OLPF and you said...

I am almost certain that that is not entirely true.

--------

I just did a deep dive into the math and, unfortunately, all chroma subsampled sensors (Bayer, Blackmagic 12K) will produce moire, even with an OLPF...

We originally were talking about using OLPFs. I suggested they are not a magic bullet. You're also made reference to still users with Fuji cameras and what not.

I get it. The idea of an OLPF is appealing. But it's a *theory* that rarely plays out that way in practice because it's really really really hard to make them to the specification required. RED use a very aggressive OLPF, which also helps with the REDCODE encoding, because it's a hell of a lot easier to compress soft detail than fine.

There's downsides to this. I mean look at how poor a windowed image is on RED cameras. The resolution you shoot at falls away dramatically visually. 2K looks like bad 720 to me. That's because you have to design an OLPF for a target resolution and RED also use a very heavy handed OLPF (likely because it helps the compression).



In reference to overall optical performance....

No, that is diffraction and proves my point. If the resolution didn't drop off like that, it would mean that the Airy Disk could get bigger and not have as deleterious of an effect on the resolution, but it, in fact, does. You can't escape the physics of it.

Have you used a lens projector before Josh ? It's really easy to see differences in lens performance based on batch to batch, focus distance, focal length. Aperture affects the image too before it ever hits the sensor. You're conflating optics with sensor resolution with a given lens.

Before you even put a lens on a camera there's a lot of things going on with the design of the lens that will affect the image, including the way it renders sharpness.


This isn't "only" diffraction. There's many factors at play here and I think you're honing in on this because you can do the maths and "see" it but it's not the whole story.

Lenses don't really have "resolution". One company started marketing their lenses as 8K lenses. It's marketing BS. A 100 year old lens can work just fine on this 12K camera and "resolve" to 12K.

Why? Let's look at some real numbers again and for that I'll use that Zeiss Batis 85mm again, since it best proves my point: https://www.opticallimits.com/sonyalphaff/1032-zeissbatis85f18?start=1 . On the A7RII using 2.5x photosite widths, that defacto number is saying that diffraction will only be experienced at an Airy Disk width of 2.5 * 4.5um = 11.25um. That means that the A7RII will be in visible diffraction at f8.4 (2.44 * 550nm * f8.4 = 11.25um.)

Josh did you read the article you yourself linked to ? When talking about the MTF of that lens with they state...

"It is somewhat odd that the resolution figures stay exceedingly high at and beyond f/8. Diffraction effects should have a more deteriorating effect here. We suspect that f/11, as displayed by the camera, is really more like f/8 in the real world (interestingly we have seen this in another Sony 85mm lens as well).
"

But you went on to argue that at even lessor stops there were diffraction issues ?

If you look at the patent, the CFA is clearly spelt out. Most aren't realising though that W pixels ALSO have ALL colour, and using colour subtraction techniques, they can also act as variable colour pixels, depending on how they're summed and combined. W pixels aren't just luminance in other words.

I'm worried you're getting a bit lost in the woods mate. It's really important to know this stuff and know what affects imaging IQ, but you can't drive every choice though this narrow (diffraction) thinking. There's lots of examples where the numbers don't really tell you the truth of what's happening.

I saw this kind of thinking on photography forums as well, especially around equivalence arguments. There's a basis in truth, but people then draw simplistic conclusions like "a 135 full frame sensor is always going to be better in low light than an MFT sensor."

You know it's standard practice for those shooting miniatures and models to work at F22 and beyond ? Do you think Nolan was worrying about lens diffraction on the Dark Knight ?

Dean Semler shot all the battle sequences in "We Were Soldiers" on long lenses at F16 because he wanted to combine long lens focal length compression and have everything in focus as much as possible. That's a creative want.

If you think always in terms of sharpness and diffraction, you're going to make some really sharp and probably pretty boring looking movies...

JB

 

[ahalpert]

If you think always in terms of sharpness and diffraction, you're going to make some really sharp and probably pretty boring looking movies...

JB

Of course, but isn't this just an academic discussion? Didn't we get here from talking about moire? (Confession: I hadn't heard of the Airy disk before.) The science of optics and sensors is inherently interesting even if it's not inherently useful.

But to be sure - some of the best cameramen have the least scientific knowledge. Trial and error rules. and ultimately making and evaluating imagery is about feeling, not thinking.

 

[Joshua Cadmium]

Josh, I appreciated the detail and depths you're going to, but remember you've already contradicted yourself. Remember when I said ALL cinema cameras can moire even with an OLPF and you said...

I did not contradict myself. You initially said "any camera can moire". Any camera includes ones with Foveon pixels. I first said there would be no moire IF we were dealing with Foveon pixels. I then said all chroma subsampled sensors (Bayer, this 12K sensor) will have color moire. Those are two different things - two different types of sensors - one needs a CFA for color and one does not.

The type that needs a CFA is going to have color moire, even with an OLPF (but can avoid luma moire with an OLPF.) A Foveon type sensor can avoid all moire with an OLPF, since color and luma resolution are the same.

I get it. The idea of an OLPF is appealing. But it's a *theory* that rarely plays out that way in practice because it's really really really hard to make them to the specification required. RED use a very aggressive OLPF, which also helps with the REDCODE encoding, because it's a hell of a lot easier to compress soft detail than fine.

There's downsides to this. I mean look at how poor a windowed image is on RED cameras. The resolution you shoot at falls away dramatically visually. 2K looks like bad 720 to me. That's because you have to design an OLPF for a target resolution and RED also use a very heavy handed OLPF (likely because it helps the compression).

Oh sure, if an OLPF is aggressive, you're throwing away too much resolution, especially at a pixel level. Plus, like I said, an OLPF is only filtering luma resolution, so you are still going to have chrome moire, unless you want to throw away a ton of total resolution.

Have you used a lens projector before Josh ? It's really easy to see differences in lens performance based on batch to batch, focus distance, focal length. Aperture affects the image too before it ever hits the sensor. You're conflating optics with sensor resolution with a given lens.

No, I am definitely not conflating things.

A lens projector is somewhat helpful to determine the resolution of the lens agnostic to the sensor you pair it with, but it is not going to tell us what we need to know about diffraction. For digital sensors, diffraction comes into play when a discrete limited measurement (photosite) tries to capture the continuous waves that a lens is seeing. That's why diffraction is measured in photosite (pixel) widths. Looking at a lens projector is like going one step backwards when trying to figure out diffraction. As soon as you introduce quantization - turning light waves into discrete, stair-stepped measurements - you limit the amount of light waves the system will see, especially as quantization (photosite size) gets smaller and smaller and can only fully contain the width of smaller and smaller Airy Disks.

The f number aperture of every lens always tells us what the diameter of the Airy Disk point of light will be (at any given wavelength of light.) And the Airy Disk gets bigger when you stop down the lens.

So, as that Airy Disk point of light diameter gets bigger from stopping down the lens, it is taking up more and more photosite widths. If an Airy Disk - the point of light - is taking up the space of multiple photosites, you simply can't resolve one Airy Disk per one photosite anymore.

To put it another way, only so many Airy Disks can fit in a given area or sensor size. They start getting so big that it takes multiple photosites to capture one Airy Disk. That's all diffraction is, and the physics of it - you can't change that. Check out this short article here: https://www.edmundoptics.com/knowle...ons-on-resolution-and-contrast-the-airy-disk/

As that page says: "As focused Airy patterns from different details on the object come close together, they begin to overlap. ... When the overlapping patterns create enough constructive interference to reduce contrast, they eventually become indistinguishable from each other. ... As pixels continue to reduce in size, this effect becomes more of an issue and eventually is very difficult to overcome. ... The smallest achievable spot size can quickly exceed the size of small pixels."

So, you have to look at the total system - lens AND sensor, if you are trying to figure out where diffraction starts.

For instance, the Arri Alexa has those massive 8.25um photosite widths. Using the standard math of the 2.5x pixel width, a modern lens on the Alexa will be in diffraction at f15.4. Sounds good. That exact same lens on the 12K will be in diffraction at f4.1. Same ratio, same lens, but the Airy Disk needs to reduce in size (which you do by opening up the lens) in order to fit into the smaller photosites.

Josh did you read the article you yourself linked to ? When talking about the MTF of that lens with they state...

"It is somewhat odd that the resolution figures stay exceedingly high at and beyond f/8. Diffraction effects should have a more deteriorating effect here. We suspect that f/11, as displayed by the camera, is really more like f/8 in the real world (interestingly we have seen this in another Sony 85mm lens as well).
"

But you went on to argue that at even lessor stops there were diffraction issues ?

If you go off of that resolution chart, the resolution peaks at f2.8. If a lens gets sharper as you stop down (with any aberrations basically going from blobs to proper points) why was it getting overall less sharp at f4? It's just diffraction - the points of light are now as perfect as they are going to be, so they just get bigger (again, Airy Disks get bigger as you stop down) and start taking up the space of more photosites. That's just the physics of it.

The reason why resolution figures stayed high while stopping down was because this lens has few aberrations. If the Airy Disk is close to a perfect point and not a blob, diffraction will have less of an effect. The reviewer just didn't know how to fully describe what he was seeing in the resolution results, because a lot of lenses are not this good.

But, if you can't get over that - throw out that review and try to find any other lens test out there that disagrees with what I've said. M4/3 lens tests, for instance, show diffraction quite soon, since the pixel sizes are smaller.

If you look at the patent, the CFA is clearly spelt out. Most aren't realising though that W pixels ALSO have ALL colour, and using colour subtraction techniques, they can also act as variable colour pixels, depending on how they're summed and combined. W pixels aren't just luminance in other words.

I was one of the first people to argue that W pixels could be contributing to color resolution - I definitely get that. I just don't know how much, if at all. It's too hard to guess without knowing the exact image kernel they are using to create a pixel based on the surrounding photosites.

I'm worried you're getting a bit lost in the woods mate. It's really important to know this stuff and know what affects imaging IQ, but you can't drive every choice though this narrow (diffraction) thinking. There's lots of examples where the numbers don't really tell you the truth of what's happening.

I saw this kind of thinking on photography forums as well, especially around equivalence arguments. There's a basis in truth, but people then draw simplistic conclusions like "a 135 full frame sensor is always going to be better in low light than an MFT sensor."

I have been doing exactly that - figuring out why a single number - 2.5x - doesn't tell me the truth of what is actually happening. I was trying to figure out why there was a difference in moire, looked at a diffraction calculator, asked myself why, and tried to see if actual data matched what the calculators were saying. They don't. Why? That ratio is too simple.

You know it's standard practice for those shooting miniatures and models to work at F22 and beyond ? Do you think Nolan was worrying about lens diffraction on the Dark Knight ?

Dean Semler shot all the battle sequences in "We Were Soldiers" on long lenses at F16 because he wanted to combine long lens focal length compression and have everything in focus as much as possible. That's a creative want.

If you think always in terms of sharpness and diffraction, you're going to make some really sharp and probably pretty boring looking movies...

First, I am absolutely not arguing that you can't stop down past diffraction. Just that it is helpful to know where it is.

I am not trying to say that everything needs to be interpreted through diffraction, if so, I'd be arguing that everyone needs to never stop down beyond f2 because why shoot the 12K if you're not getting 12K worth of resolution. I am absolutely not saying that. I could also say you need to always stop down to f5.6 and beyond in order to limit chroma moire. I'm not saying that either.

If one doesn't care about diffraction either way and it's more of a hindrance, just don't think about it. I find it helpful, but not everyone will. My goal was to determine why there were differences in moire and I think we are closer to understanding why. I'd rather know the reasons than have it remain a mystery.

Also, I definitely agree with you that resolution is not the end all be all. I keep saying that. I personally care way more about color reproduction than anything else, which is why I was drawn to this camera, not because it was 12K. I'd much prefer to be shooting at 2.8K Alexa than almost anything else out there. But for other people, resolution may be more important than any other criteria.

Also, for your examples, film has a much lower resolution than what we are talking about, so you can stop down quite a bit more before hitting diffraction. (And to determine diffraction on film, find the resolution of a given piece of film. Take the width of that film in mm, divide by resolution and multiply by 2.5x - or a smaller ratio.)



--------



One last thing I wanted to mention, and it will really help with what I am arguing - you will start to see the effects of diffraction at a per pixel level. If you're not going to be viewing things at a per pixel level, then it doesn't matter as much. If one's plan is to take 12K worth of resolution and smoosh it into 4K worth of delivery pixels, you are already going to be blurring all of those pixels together anyways, so diffraction is not going to be seen unless you really stopped down the lens.

For instance, let's say you stopped down and lost 20% of the total resolution you would otherwise get. You'd still be at about 10K. That still gives you a 25% oversample when viewing at 8K.

However, if you are going to be punching in a lot, or you are going to be shooting with a smaller crop of the sensor, then diffraction will have much more of an effect. Does that help more with what I am saying?

 

[John Brawley]

But to be sure - some of the best cameramen have the least scientific knowledge. Trial and error rules. and ultimately making and evaluating imagery is about feeling, not thinking.

For sure.

I like to be aware of the imaging “science” but I also get nervous when I see absolutism, especially around numbers that in my experience, get turned into “rules” and I think rules should really be avoided in creative processes.

There’s plenty of other reasons Moire (to bring it back to topic) could have occurred as well on the footage in question.

And for whatever reason, It’s not been a show stopper on any of my 12K shooting so far. And I’ve been shooting right in the sweet spot of non-diffracted super high performing lenses.....so....

JB

 

[Joshua Cadmium]

But to be sure - some of the best cameramen have the least scientific knowledge. Trial and error rules. and ultimately making and evaluating imagery is about feeling, not thinking.

Oh, sure, you don't need to know any deep knowledge in order to make aesthetically pleasing images or serve a story. But it can be helpful and keep you from doing too much trial and error, especially the error part. You want the technical aspects to inform you up to a point and they fade away so you can focus on the creative side. But they go hand in hand.

I like to be aware of the imaging “science” but I also get nervous when I see absolutism, especially around numbers that in my experience, get turned into “rules” and I think rules should really be avoided in creative processes.

There is quite a bit of pseudoscience in the photography / cinematography world. I think part of the reason is because the physics and math start to get real wonky real fast and it's easy just to make shortcuts or rules. But like any rules, you usually need to know why they are there before you can effectively ignore or break them.

 

[Joshua Cadmium]

One more thing I wanted to mention about using 2.5x pixel width for diffraction. The calculators online and what I have been using is based on light being at green (usually 550nm) since that is right in the middle of the spectrum. But just using green doesn't tell the whole picture.

Violet light (just at the edge of UV) is around 400nm, which means it actually creates a smaller Airy Disk at every f stop. So, using that 2.5x number on the 12K (a width of 5.5um) but with 400nm light, we get violet light diffraction right at f5.6. That's a much deeper f stop than the f4.1 you get when running the math at 550nm. [The math, once again, is 5.5um = 2.44 * 400nm * f stop. OR f stop = 5.5um / 2.44 / .4um]

Far red light is around 700nm, which means it creates a larger Airy Disk at every f stop. The 2.5x math puts far red light diffraction at f3.2.

Warmer tones are going to experience diffraction before cooler tones do. That's just the physics of light.


For comparison at 2.5x width:

400nm = f5.64
550nm = f4.10
700nm = f3.22

For comparison at 1.5x width:

400nm = f3.38
550nm = f2.46
700nm = f1.93

For comparison at 1x width:

400nm = f2.25
550nm = f1.64
700nm = f1.29


Is this helpful? Maybe not to everyone, but I certainly find it absolutely fascinating. It also again shows why that 2.5x defacto standard for diffraction is just a simplistic number like any other rule.

 

[Guest]

Warmer tones are going to experience diffraction before cooler tones do. That's just the physics of light.

Well, they were getting moiré on the skate boarder's Khakis.

 

[Guest]

For comparison at 2.5x width:

400nm = f5.64
550nm = f4.10
700nm = f3.22

For comparison at 1.5x width:

400nm = f3.38
550nm = f2.46
700nm = f1.93

For comparison at 1x width:

400nm = f2.25
550nm = f1.64
700nm = f1.29


Is this helpful? Maybe not to everyone, but I certainly find it absolutely fascinating. It also again shows why that 2.5x defacto standard for diffraction is just a simplistic number like any other rule.

Thank you. Yes, I find it helpful.

Could also be worth saying that since most lenses aren't their best across the whole image, that framing subjects not in the center could also be part of the reason why one person might see an issue, and not another person.


I recall Bradford Young, in an interview talking about shooting 'Solo' on those crappy~ Arri DNA lenses, and he never used to center his subjects, much preferring rule of thirds and placing a subject to the left or right. But the DNA lenses wide open were so soft particularly in the edges, he found himself switching up how he composed an image, playing the subject dead center. Something he used to think was bad.


Yes, we can fall in love with the flaws and the happy accidents, or improvise, or feel it, but then someone like Roger Deakins or Chivo use some lenses that a lot of my friends find unappealing. I showed a director friend some Arri signatures, and he was so unimpressed. Yet, he loved '1912', so, go figure. There are the rules, breaking the rules, and using the rules to make it look like you broke the rules.... to... hahah. and so on.

But thanks again, Joshua, those numbers were never really any of my concern, but I can use them if they come in handy in the future. If nothing else, it was entertaining and informative read. thanks/

 

[Guest]

Black Magic, as far as camera companies go, seems like a good company. If they have any fault, they are just too cheap, and that means they never really give a premium or particularly confidence inspiring experience for pros. However, with how fast technology advances, seems to be a little better business model to sacrifice quality and durability for better IQ and lower cost. In many ways, or at least for most creatives trying to make a living outside the corporate world of film and television, this is many times better than RED. Red, to me, has better cameras, but marginally... and really only subjectively. There is more to a camera than one spec here or there. Also, RED is not much better than Black magic, since they are just repackaging off the shelf technology and branding it as premium, and charging a butt load for it. I seriously don't care what any of these companies charge for whatever. They can do what they gotta do, but I sure do wish one company other than Arri would make a camera that makes sense and has the IQ. In the 2 plus decades of high end video, there have been so few cameras that combine sensibility, quality, and iq. I think Sony and Canon have made a few sensible things here and there, Panasonic definitely has, but those models generally suffered in iq. Ever since the 1DC, I realized how amazingly easy it would have been for Canon to slap a 10bit higher bit rate codec on their s35 HD crop, or hell even give the 1DC a 2K crop as well! The rest of the camera was sensibly laid out and weather sealing was great, and it was a tank. I dropped the thing and it landed on the weakest part and still had no issues. Canon service was amazing too.

Now, cameras are getting better. 2020 has shown that even when these companies are still being very conservative and stratifying their products, they cannot help but to make some amazing things. Why did they never give us 10bit 2K? Dog knows. But that is in the past. They went from 17Mbps HD to 12K raw in the blink of an eye, but it was amazing how fast they skipped over good 2K in favour of passable 4K. And now, 6 and 8 K raw is everywhere and it feels like no one in particular asked for it. haha.



So, despite maybe Blackmagic not being "my" brand, one could say they've done right by their user base. I prefer to pay more and get some premium features. Just that the gap between BM and Arri is waaayy too big. And Canon held back their stills division while their Cine EOS floundered about finding its legs in the actual "cine" aspects, before embracing their "cine-like" aesthetic, which is admittedly still capable of enabling user to capture gorgeous images. And the original C300 felt like a toy and ergonomics were weird. Luckily it looked professional enough, and provided some advantages over other systems at the time (larger sensor, faster RS, BNC, cheap media, XLR, many cheap lenses, built in ND, etc, etc).

I say all that, because, when I talk about this 12K sensor, I am trying to just talk about this sensor, right now. Separate from its brand for a moment. It is an interesting sensor, because it pits the technology against itself. And it makes me wonder were we are going with cameras. Everything was so lame from 2004 to 2019. As soon as HD made the DVX100 obsolete, it has been one awkward goose of a product after another. Yes, a few expensive gems, and a few "bargains of IQ", or happy accidents in ergonomics, came along. But by and large it felt like a mess of cameras made by marketers and focus groups. "Does this look like an HD camera to you?" "Would more K's make you upgrade if everyone demanded more K's?"

So, this 12k thing:
- Does it shoot 12K? Well, technically yes, but... you reeeeally don't want to crop in on all that noise.
- Does it resolve 12K or downsample to 8K nicely. Yes, it is sharper than RED's 8K, but it will moiré. The R5 might be equally sharp though.
- Does oversampling resolution allow for better low light performance? No. However, the CFA does allow for better than expected noise performance, (insert Leo Decaprio meme from D'Jango)
- Does it have any advantages in dynamic range or latitude. No.
- Are the colours better? That is subjective. Technically yes, practically, not in the shadows or the highlights, but technically, in the midtones, there is more tonal and hue information.

So, what are the advantages? "It's only $10K, and it has a flippy screen."


I can see this camera being amazing on a soundstage for VFX and fashion work. And even for available light shooters out in the wild, it gives plenty of IQ and sensitivity. THERE ARE NO REAL DISADVANTAGES to the BM12k, that I can see, or have heard about. My question is just, "is it better?" Will have to see how the A7sIII raw out and Arri's s35 4K sensor play out. I think BM has made a very interesting and unique camera that has some good things going for it. It certainly isn't particularly lacking. But I have seen the Monstro in raw, and I am not certain the BM12K will be replacing either the Mini LF or Monstro, and those cameras' origins predate the BM12K by a few years. So, the BM is still more of a bargain version, rather than a 1:1 replacement. Believe me, I wish someone would dethrone RED's over priced toys. But I don't begrudge RED for being over priced if no one is definitively challenging Arri. Sony landed the Venice, and that camera does somethings that put RED in 3rd place, for sure, but a Venice package can still cost near Arri prices without being definitively better than Arri, and people are getting RED packages for 20-30 grand cheaper than a Venice one, and Sony's next product down is the Fx9.... so......

But seeing as the 12K sensor is pushing the limits (not in all good or bad ways) of CMOS tech, where are we going in terms of image quality for the future? Seems we have plateaued, other than a few refinements and denoise tricks. We are still balancing the same tradeoffs since 2008. Basically that means, whatever camera you have now is going to be good enough for the foreseeable future, until either they come out with something better than CMOS, or if everyone starts doing good DGO. (is it expensive or just an Arri patent? I know Red wouldn't do it, because it isn't off the shelf hardware yet, lol).

So yeah, 2020 has been both the one of the biggest advancements in the general quality of available cameras, while also just being more of the same. Many of us have had access to similar IQ for a long while (albeit at higher price points). So, that is good for those that already have a nice camera, as it means no need to upgrade, and it is also nice for those buying their first camera, as they don't have to wade through figuring out which camera is actually good... they all are now.


Should be interesting to see "smart" features added to cameras in the future. Like imagine having F35 level IQ, but some very smart filters that work with AI to "fix" things in camera. Instead of face filters, the face tracking AI implements more subtle enhancements. Other than that, I mean, it seems like CMOS has run its course. It gave us cheaper cameras, that run longer on batteries and require less cooling to run than CCD. So, gotta say thank you to CMOS. But other than a DGO camera introduced back in 2010, it really never fully devilered on what matters to cinematographers.


To be honest, brand loyalty in 2020 shocks me. But, I still have to give it up to all the engineers that made it possible to shoot really nice video. the fact that these cameras even exist as they are still amazes me. Even if they never really got to where I had hoped (actually expecting the Arri 4K sensor to be the one, but who knows). I think I have bought my last CMOS camera, and happy to let that one ride out. Idk, maybe if a lightweight highly ergonomic master piece comes out, but the chances of that are highly unlikely. Good ergonomics typically takes a back seat to technological advancements.

 

[Tom Roper]

 

[Guest]

That 2K crop looks really clean actually, and no moire. pretty good!