Wishful thinking: Nikon FX mirrorless to sport square sensor

[BeBu Lamar]

- I appreciate that Oskar wanted to make a larger picture area available, but it could just as easily have been made 24 by 32 mm in size. I suspect he just happened to have some cine sprocket rollers handy, and couldn't be bothered to gear them down sufficiently for the wind-on mechanism.

 

We no longer have to fit our frame into the sprocket width of cheap cine film, and it would be easy to use a frame size of, say, 27x36mm, or even 25 x 35mm. This gives us a 1.4:1 aspect ratio that fits within the same image circle as full-frame, and also fits ISO paper sizes almost perfectly.

 

Anyway, the point I was trying to make is: That no matter how ungainly, costly or awkward to compose into, any ill-conceived format size will be seen as 'perfect' and sacrosanct by some faction or other. Like, say, the ill-informed, money-driven members of the 16:9 consortium that drag up the golden ratio (nowhere near 16/9, and never meant to be used for an aspect ratio) as an excuse to force us to watch our entertainment through an unnatural letterbox.

 

As for the 16:9 aspect ratio I believe it was chosen as a video format that is a good compromise between various video formats from 4:3 to 2.35:1. For still it's a lousy aspect ratio.

[chulster]

Why then record the entire square and fill the memory card with data that's not needed? And if you think you may need it - then why go through the trouble to select the desired aspect ratio in camera since apparently you may not stick with your in-camera choice?

 

For JPEGs, and because you'll want to see the cropped preview in the viewfinder for easier composition.

 

There will, of course, be a setting to decide whether to commit the entire sensor data to file or only the cropped-in part.

[Gary Naka]

I thought the 16:9 format was trying to get closer to the wide-screen theater format, without horizontal compression of the image.

[ben_hutcherson]

Why then record the entire square and fill the memory card with data that's not needed? And if you think you may need it - then why go through the trouble to select the desired aspect ratio in camera since apparently you may not stick with your in-camera choice?

 

The concept is oddly reminiscent of APS film.

 

Every frame is recorded as APS-H(16.7x30.2mm). If you select a different crop in camera(APS-C, which is 16.7x25.1, or APS-P, which is 9.7x30.2) the frame is "tagged" either optically(in cheap cameras) or magnetically(in better cameras) to tell the printer to print in the selected aspect ratio.

 

On the rare occasion I shoot APS film, I tend to compose using the viewfinder APS-C indications, but always leave my camera set to the rather unwieldy 4:7 APS-H format. If setting to APS-C would save film, I would use it, but it doesn't.

[chulster]

The concept is oddly reminiscent of APS film.

 

Every frame is recorded as APS-H(16.7x30.2mm). If you select a different crop in camera(APS-C, which is 16.7x25.1, or APS-P, which is 9.7x30.2) the frame is "tagged" either optically(in cheap cameras) or magnetically(in better cameras) to tell the printer to print in the selected aspect ratio.

 

On the rare occasion I shoot APS film, I tend to compose using the viewfinder APS-C indications, but always leave my camera set to the rather unwieldy 4:7 APS-H format. If setting to APS-C would save film, I would use it, but it doesn't.

 

Sehr interessant!

[glen_h]

As I once knew it, the Leica was designed to preview movie frames in a larger format.

 

It seems to me, then, that it should have had the same 4:3 as movie frames (at the time).

 

Otherwise, it would be interesting to design a sensor with a cross shaped array that one could select either 36x24 or 24x36.

(Or, as noted above for APS, store the whole thing and decide later.)

 

As noted, many Nikon lenses have masks, but as I understand it the discussion is for a new set of lenses.

They could take the existing design, remove the mask, change the mount, and generate a new series.

[donbright]

Or it will be one of two camera's offered one being 44X33 sensor size, would justify the large mount.

[BeBu Lamar]

Large mount and shallow flange is good for using on a view camera and use shift and stitch to create large format images.

[Andrew Garrard]

Ah, aspect ratios. I've had articles published on this. Fortunately others made most of my points before I could rant properly.

 

3:2 is a pretty good compromise for wasted area between the range of common aspect ratios - especially with the prevalence of viewing on 16:9 screens. Anything else assumes that rotating the camera is so inconvenient that it's worth the cost and weight of a larger sensor and lenses. Panasonic do have multi-aspect cameras, but also smaller sensors. If this is really an issue, a portrait grip on the camera is a much simpler solution.

 

16:9, by the way, was a compromise between 2.35:1 cinema and old 4:3 televisions. And then people started broadcasting 14:9 content as a compromise between 16:9 and 4:3. Fortunately that stopped.

 

A while back, many computer screens were 16:10. Easier to ship than a squarer format (and you can advertise the diagonal to make it sound bigger), and arguably worse for fitting lines of text on the screen. Fortunately it was also half a common 5:4 screen, which was nice for multi-monitor rigs. 16:9 is just dumb. It fits very few integer resolutions (especially with large powers of two that computers like). Case in point, 1080 vertical pixels doesn't divide by 16 - but the chroma channel of basically every HDTV broadcast has to be a multiple of 16 pixels. So every HD signal wastes a little bandwidth on eight rows of pixels you can't see. Don't get me started on 1366x768...

 

Anyway. 3:2 is fine. I don't hate 4:3. Sqrt(2) would make some equivalence discussions easier, but it's probably not worth the weird numbers. 2:1 isn't awful, but probably going too far. But not square!

[glen_h]

As far as I know, they interpolate the pixels.

 

The 37 inch TV set I have, I believe has a 1366x768 LCD display.

 

Obviously that doesn't agree with either 720p or 1080i, so both need some interpolation.

 

But if I connect up the VGA port, or HDMI to a computer, I believe it figures out 1366x768.

[Andrew Garrard]

As far as I know, they interpolate the pixels.

 

The 37 inch TV set I have, I believe has a 1366x768 LCD display.

 

Obviously that doesn't agree with either 720p or 1080i, so both need some interpolation.

 

Indeed. I suspect this was a cynical attempt to get people to upgrade to 1080 panels because they looked much sharper. Which they wouldn't have done to the same extent if the industry had just sold 1280x720 televisions without scaling 720p content. Or I could just be an old cynic.

 

If you want to do really nice scaling, one way is to cover the image to Fourier space and clamp the frequencies before rescaling. One efficient way to apply a Fourier transform involves factoring the resolution into hopefully small prime numbers. 1366? 2x683. 683 is a fairly large, considering, prime number. 1366x768 is about a stupid a panel as they could have made, with the dubious excuse of repurposing 1280x768 production lines.

 

But if I connect up the VGA port, or HDMI to a computer, I believe it figures out 1366x768.

 

It does - EDID should report the native resolution. Although I could go on a separate rant about how the first 1080 televisions could only do interlace, or p30, or would always scale the image despite having enough pixels, or hood some of the pixels behind the bezel. I had some frustrating times with video drivers in the 2000s.

[glen_h]

In the 1D (digital audio for example) case, it is usual to use an IIR filter.

The math isn't too bad, and it can be done while streaming.

 

The same math works in 2D, but takes much more calculation.

Also, the data rate is much higher for video data than audio, so, as far

as I know, no-one does the appropriate frequency filtering for the actual

interpolation. They cheat, and it is close enough for most of us.

[Andrew Garrard]

Indeed - fortunately the transform in 2D is separable. It's normal to use a FIR, but I've seen visible differences in rescaling video up to thirteen nodes of a sync. You're lucky if most video scalers use 5 or 7. Somewhere around 16x16 filters it becomes more efficient to do everything in Fourier space. Although not so much if one dimension is 1366...

 

(To be fair, I was doing this "properly" for educational reasons - I needed to separate the frequencies. But it's still an annoying resolution.)

[chulster]

Objections to the square sensor fall mainly into two groups: economic and ergonomic. The economic objections can be dispensed with by shifting the time frame, from the present, to a future time when sensors are so cheaply produced that the cost difference between a given sensor size and one that is 50% larger is negligible—just as the material (if not retail) cost difference between a liter of Coke and 1.5 liters is negligible. We can imagine that, in this future, silicon wafer manufacturing has gotten so advanced that the difference in yield, or defect rate, between the larger and smaller sensors is also negligible, so that manufacturers have no economic reason not to equip cameras with square sensors.

 

Naturally, there will be objections to this time shift. It may be that, by the time the economics of CMOS sensor manufacture cease to make a significant difference to the cost of a camera, some new imaging technology will have come about, rendering the above scenario moot. Well, this whole debate is moot, so who cares.

 

Now, on to the ergonomic quibbles, which are, mainly, that the camera and lenses will need to be bigger/heavier in order to support the larger, square sensor. I grant that the body will need to be 12mm taller just to fit the taller sensor—unless Nikon can find a way to shrink the parts of the camera surrounding the sensor so that there is no net increase.

 

Regarding the lenses, though. What if I told you that, if you only want to shoot in the 3:2 aspect ratio, you could buy a smaller, lighter lens suited to that frame and use it on this same camera?

 

There is no reason why every lens for a square-sensored camera must project an image circle large enough to cover the entire sensor. For those who only love 3:2 (or 4:3, or whatever), lens makers can design and produce smaller, lighter lenses that only cover a 3:2 crop of the sensor. Of course, use of such a lens would preclude the major benefit of the square sensor, which is the ability to select an aspect ratio in the viewfinder and, optionally, to change it in post without quality loss. But if size and weight reduction matter more to you, use a smaller lens! The lens's electronics could even tell the camera that the lens is optimized for a certain aspect ratio, and the camera could then make other ratios unavailable for selection. It would be the best of both worlds.

[BeBu Lamar]

For camera sensor I would prefer a 4:3 or 5:4 aspect ratio but for computer monitor a 1:1 is best. With the square screen there is no need to rotate it like I do now with my 4:3 screens. With a large square screen you can put the tools and menu on the side or on the top depending on whether you are working on a portrait or landscape document.

[Andrew Garrard]

Chulkim: you're effectively suggesting that FX lenses be like DX now, supporting a crop of the maximum mount? Interesting, but I don't see big sensors getting all that cheap (they don't, after all, obey Moore's law) and a bigger shutter has its own costs and difficulties. Currently that also applies to the mirror and prism, but obviously not for mirrorless.

 

So it's possible, but I maintain it's a lot of effort to go to just to avoid turning the camera through 90 degrees and using a grip. And in mirrorless, the shirt assembly could just rotate in the camera. (I'm assuming the concern is wasted image area for portrait images. If you actually just want square images, I could make similar objections about why the sensor is a poor fit for wide panoramas.)

[Gary Naka]

For camera sensor I would prefer a 4:3 or 5:4 aspect ratio but for computer monitor a 1:1 is best. With the square screen there is no need to rotate it like I do now with my 4:3 screens. With a large square screen you can put the tools and menu on the side or on the top depending on whether you are working on a portrait or landscape document.

 

Some of the OLD text based word processors had a vertical screen, so the operator could see the entire page, rather than only a part of the page. And that did work well.

 

Today I get a LARGE WIDE screen monitor, so that I can work on multiple windows/applications at the same time.

In fact if I had the room I would setup TWO monitors, for even greater horizontal workspace, which makes it easier when I have to cut from one doc/application into another. But this goes back to how I like to work.

I am also a spreadsheet guy, so I like WIDE monitors to give me more horizontal visibility of my spreadsheets, which goes out MANY columns.

[BeBu Lamar]

Some of the OLD text based word processors had a vertical screen, so the operatvor could see the entire page, rather than only a part of the page. And that did work well.

 

Today I get a LARGE WIDE screen monitor, so that I can work on multiple windows/applications at the same time.

In fact if I had the room I would setup TWO monitors, for even greater horizontal workspace, which makes it easier when I have to cut from one doc/application into another. But this goes back to how I like to work.

I am also a spreadsheet guy, so I like WIDE monitors to give me more horizontal visibility of my spreadsheets, which goes out MANY columns.

 

I currently have 2 monitor 4:3 and they are in portrait orientation. My speadsheet generally has more rows than columns.

[Andrew Garrard]

And I normally run three monitors at work, two 16:9 and a 16:10. They're all in landscape, largely because it's easier to configure and not all the monitors available to me can rotate. I don't have VESA mounts, so they're necessarily horizontally oriented. I do have a lot of windows open, but I also do a lot of scrolling.

 

Somewhat taller monitors would btw a bit better, although at some point I have to see over them. Maybe VR will eventually have a pixel density that will let us choose, although I've been a cynic about the technology for a long time now.

[Gary Naka]

I currently have 2 monitor 4:3 and they are in portrait orientation. My speadsheet generally has more rows than columns.

 

More rows than columns . . .

You don't qualify as a spreadsheet "junkie" :D

 

Well, then it depends on how many rows.

[BeBu Lamar]

More rows than columns . . .

You don't qualify as a spreadsheet "junkie" :D

 

Well, then it depends on how many rows.

My spreadsheet sometimes has quite a large number of column that I can't display them all in one screen even with a 4K display. However the number of column sometimes in the ten thousands.

[chulster]

I wish to withdraw my advocacy of the square sensor, in favor of the circular sensor. Everything the square can do, the circle can do better.

[rodeo_joe1]

I wish to withdraw my advocacy of the square sensor, in favor of the circular sensor. Everything the square can do, the circle can do better.

 

- Only if wasting unbelievable amounts of silicon wafer space is 'better'.

[chulster]

- Only if wasting unbelievable amounts of silicon wafer space is 'better'.

 

Oh, but you must get out of that efficiency mindset! Your desktop computer's CPU probably spends >90% of its time at 10% utilization or less. Did you choose the slowest available processor in your last computer therefore?

[Andrew Garrard]

I wish to withdraw my advocacy of the square sensor, in favor of the circular sensor. Everything the square can do, the circle can do better.

 

Tessellate? (Hexagonal sensors ftw!)

 

Oh, but you must get out of that efficiency mindset! Your desktop computer's CPU probably spends >90% of its time at 10% utilization or less. Did you choose the slowest available processor in your last computer therefore?

 

I hope people might trade cost against how well the job gets done...