What would the Ideal MP be for a FF 1:1:1 Foveon DSLR?

What would the Ideal MP be for a FF 1:1:1 Foveon DSLR?


  • Total voters
    0
O.K. I thought I would show some results of practical observation. I picked two cameras that seemed to produce similar image quality, though their sensor size is vastly different . . . the Nikon V3 and the Sony A7s. Yes, the Nikon captures more pixels, but that is the point, isn't it? We're trying to see if there will be an advantage to more pixels packed closer together. I think this demonstrates it well.



ISO 200
ISO 200



ISO 800
ISO 800



ISO 3200
ISO 3200

As we can see here, what is happening is the dramatic difference of added noise at higher ISO vs. with a small sensor. What happens to the image from the larger sensor is a much-less-dramatic change in "resolution" or ability to capture detail.

THIS exemplifies why large pixels are important, but primarily at high ISO rages, because other than that, we don't see a huge advantage to large pixels. The D800 proved that you can put 3 times as many pixels on a sensor, yet still get a greater dynamic range at ISO 200. The D700 did NOT have the same high dynamic range of the D800, so pixel size obviously has NOTHING to do with dynamic range at lower ISO ranges. Noise? Well, we can see what happens with small pixels in the noise department by the top image . . . there is practically no difference from the tiny pixels of the V3 sensor to the huge pixels of the Sony A7s sensor (at ISO 200).

So it is not until high ISO settings that we experience image degradation based on pixel (photo-site) size. Since I usually shoot at ISO 100 anyway, I don't see a reason to buy camera with a huge sensor or big photo-sites (pixels). For me, the advantage of the small APS-C sensor is the dominant factor.

Now for an even bigger difference:



Nikon D700 vs. point and shoot cameras.
Nikon D700 vs. point and shoot cameras.

I was really surprised to see the results of the D700 vs. point-and-shoot cameras at low ISO. Some of these cameras even have serious zoom capability. One is even waterproof (Nikon AW110). ALL these point-and-shoot cameras have tiny sensors - the largest is the sensor in the Canon, which is 1/1.7" size (much smaller than what the Sony RX100 series or Nikon 1 series cameras have in them).
 
I would say 36mp, as I see nothing wrong with the pixel size of the Merrill. However maybe 24mp would make the camera usable above ISO 400? Also that's a good size image and it would be more forgiving on lenses and technique. So I voted 24.
Jack do you realize most people consider the Merrill capable of good quality results at ISO 800? I think the Quattro DEFINITELY is capable of good results at ISO 800 and possibly even at ISO 1600 (depending on the camera, image processing, etc.). If you were to double the size of the sensor, you could make a Quattro with double the pixels (39 MP) and presumably get similar per-pixel results. Of course, as an overall image, at high ISO a 39 MP image will look better than one with similar per-pixel noise characteristics, so maybe ISO 3200 would be semi-usable with a full-frame Quattro with a 39 MP top layer. Then there are the other advances that will be made in the next year or two, which I believe will make ISO 3200 more than just semi-usable with such a high resolution full-frame sensor. We are not limited so much by the physical size of the photo-sites as much as we are limited by the design of the components of the sensors, the software and firmware involved, etc. I think that when the Canon 7 D came on the market, with 50% more pixels than the 5 D, yet it produced less noise (or about the same) at ISO 3200, we all should have learned that pixel size (photo-site size) is not the key ingredient of even high-ISO performance.
 
I would say 36mp, as I see nothing wrong with the pixel size of the Merrill. However maybe 24mp would make the camera usable above ISO 400? Also that's a good size image and it would be more forgiving on lenses and technique. So I voted 24.
Thanks Jack. 24MP does appear to be the most popular of the poll choices. Comes out as about 6um pixel pitch, 6000x4000px, "B" size (A3) print at 360 dpi.
 
Last edited:
Ted it looks like there are about as many votes for the 36 MP sensor as a 24 MP sensor. I guess maybe a 28 MP sensor might be a good compromise. If that were to happen, the equivalent would be approximately 56 MP (in Bayer pattern CFA land). A 28 MP (x3) full-frame Merrill sensor would no doubt match or exceed the image quality (low ISO "resolution") of a Pentax 645 Z, and come very close to capturing image quality as good as a 60 MP Hasselblad (at ISO 100 and 200). THAT would be GREAT!

Frankly, I think that the 24 MP choice is not enough . . . with a 48 MP equivalent. If they were to introduce it right now, it would be, but we have to be forward-thinking. In a year or two, when it COULD come on the market, 36 MP cameras will be old, and something higher (like 48 MP will be right around the corner). I'd like to see Sigma do more than just match the new cameras that are coming on the market. I'd like to see Sigma make a camera (in secret) that will be ahead of its time, as far as resolution . . . just like they did with the SD1. They introduced the SD1 LONG before they started selling it, and they crippled the sales with a ridiculously high price. I think that they could make a full-frame camera in secret, and introduce it just a month or two before they are ready to start selling it. THAT would be awesome, because it wouldn't give companies like Sony and Nikon much time to respond, before they start selling the full-frame beast.
 
Ted it looks like there are about as many votes for the 36 MP sensor as a 24 MP sensor.
Not as of right now, and not when I looked yesterday.
 
I voted for More! Theoretically more (ideal) pixels is always better than less pixels - you can use better processing algorithms then.

Considering required in-camera processing power however, then about 16-24MP seems optimal. Enough data for image processing and not that big need for power (and storage).
 
I voted for More! Theoretically more (ideal) pixels is always better than less pixels - you can use better processing algorithms then.

Considering required in-camera processing power however, then about 16-24MP seems optimal. Enough data for image processing and not that big need for power (and storage).
 
Ted it looks like there are about as many votes for the 36 MP sensor as a 24 MP sensor.
Not as of right now, and not when I looked yesterday.

--
Cheers,
Ted
What do you mean Ted? I see 12 votes for 24 MP and 10 votes for 36 MP. Isn't that pretty close? When I wrote my comment I think it was 10 and 9.
Quite right. In fact I said as much yesterday: http://www.dpreview.com/forums/post/54292497

And, yes, that it 'pretty close' - good enough for govermint work, eh.

Today, I was looking at it skewiffy.

--
Glurk,
Ted
 
Last edited:
My 36mp vote is based on 36mp Q configuration that can do 9mp 1:1:1 super pixels and will be very good at ISO1600. Why bother with more pixels when less will do?

Sigma needs to iron out some of the issues with Q, namely colour blotches in shadows and limited DR.

--
Maple
 
Last edited:
Having experienced the F17 and the Merrill imagers, IMHO there is something magical about the 7.8u geometries found in the SD14, DP1, DP2. While the Merrill offers insane resolution, the F17 seems to offer the must natural, and film-like, rendering.

Just ordered a DP2Q, so can't comment, but excited to learn about her rendering characteristics, and appreciate all of the thoughtful contributions around highlight recovery at ISO 100 vs. 200, over-stated ISO calibration, AFE amplification stepping intervals, etc. ..what a terrific forum!

My vote is for 1:1:1 FF with large, 7.8u pitch and improved fill factor, an advanced Foveon-optimized image processing pipeline, not one made of Bayer parts like Fujitsu's TRUE engine, and most importantly, an Adobe compatible workflow. The larger photo-sites stave off diffraction limits so one can shoot happily at F16, which I would presume would be important to anyone who is requesting a F135 format, as they would want access to the full gamut of DOF possibilities, especially if one expects to realize gains when outputting at 760 dpi shooting stopped down past F8.

Great poll Ted, thx!
 
On an Epson printer, an image looks great at 180 dpi, better at 360 dpi, even better at 720 dpi, etc.

The more information I can provide to a fixed-dimension output space (a print as opposed to a monitor), the better. Anyone who claims otherwise is just settling for "good enough."

The print is all about information: as much information as possible processed correctly.

Punkt!
Quite right and maybe the forums (all of them actually) should be split into those that display the final image by print and those that like to look at 100% on a monitor.

Different beast all together. The Epson printers are based on a 1440 weave pattern so any submultiple of 1440 will work better since it will require one less resampling. For B&W I use Quadtone Rip (QTR) and have found the best setting for at 360 or 720 DPI and a 1440 weave pattern in the RIP.

If I print color - which is only when my wife nags me enough - I use the Epson driver and a 360 DPI.

--

Truman
www.pbase.com/tprevatt
 
Scott . . .


Diffraction is caused by the aperture, so that is a limit of physics, but why not allow for greater resolution, when that aperture is larger? Sometimes people shoot at f5.6 or even f4, where a sensor with greater resolution would be useful. You can always bin to half resolution (2x2), and make a 39 MP sensor output equivalent to a slightly less than 10 MP output. If you're worried about the amount of data, why? The 14.7 MP per layer Merrill sensor is almost half a 39 MP sensor and more than half a 28 MP per layer sensor. Processing speed has more than doubled since the Merrill senosor was introduced, and with a greater fill factor, a 28 MP sensor should be able to produce better per-pixel image quality than the Merrill. I predict that the Merrill per-pixel image quality is achievable in a 39 MP per layer full-frame sensor, with today's technology, and with developments over the next year or two, a full-frame sensor could be made to perform at least as well, even though it offers MUCH higher resolution than a 28 MP per layer sensor would. I understand the desire to not go too far, but Bayer CFA sensors are advancing, and we are likely to see another big jump in resolution to 48 MP or more, by the time a full-frame Foveon sensor can be released into the market in a camera body. It is important that it offers some advantage over a 48 MP or higher resolution Bayer pattern CFA sensor, or it will not be practical to make it. Foveon sensors will apparently NEVER equal CFA in high-ISO noise quality, so why continue to make efforts toward that goal?

My suggestion is a stop-gap step into full-frame . . . a 28 MP compromise, which should be competitive with a 48 MP full-frame Bayer pattern CFA sensor and a step toward a true 39 MP (8K) resolution, which could be achieved with a full-frame Quattro sensor. That's my thinking. Diffraction will ALWAYS be a limit . . . at some aperture . . . no matter what resolution your sensor is.
 
The World is going digital. Paper prints will be a thing of the past soon. 4K is here now. 8K is coming. Keep that in mind.
 
On an Epson printer, an image looks great at 180 dpi, better at 360 dpi, even better at 720 dpi, etc.

The more information I can provide to a fixed-dimension output space (a print as opposed to a monitor), the better. Anyone who claims otherwise is just settling for "good enough."

The print is all about information: as much information as possible processed correctly.

Punkt!
Quite right and maybe the forums (all of them actually) should be split into those that display the final image by print and those that like to look at 100% on a monitor.

Different beast all together. The Epson printers are based on a 1440 weave pattern so any submultiple of 1440 will work better since it will require one less resampling. For B&W I use Quadtone Rip (QTR) and have found the best setting for at 360 or 720 DPI and a 1440 weave pattern in the RIP.

If I print color - which is only when my wife nags me enough - I use the Epson driver and a 360 DPI.
Sounds like you two gentlemen don't factor the human visual system and its acuity into your needs?
 
On an Epson printer, an image looks great at 180 dpi, better at 360 dpi, even better at 720 dpi, etc.

The more information I can provide to a fixed-dimension output space (a print as opposed to a monitor), the better. Anyone who claims otherwise is just settling for "good enough."

The print is all about information: as much information as possible processed correctly.

Punkt!
Quite right and maybe the forums (all of them actually) should be split into those that display the final image by print and those that like to look at 100% on a monitor.

Different beast all together. The Epson printers are based on a 1440 weave pattern so any submultiple of 1440 will work better since it will require one less resampling. For B&W I use Quadtone Rip (QTR) and have found the best setting for at 360 or 720 DPI and a 1440 weave pattern in the RIP.

If I print color - which is only when my wife nags me enough - I use the Epson driver and a 360 DPI.
Sounds like you two gentlemen don't factor the human visual system and its acuity into your needs?
Can't speak for Laurence, however, in my case my workflow is judged by what is best hanging on the wall.

 
On an Epson printer, an image looks great at 180 dpi, better at 360 dpi, even better at 720 dpi, etc.

The more information I can provide to a fixed-dimension output space (a print as opposed to a monitor), the better. Anyone who claims otherwise is just settling for "good enough."

The print is all about information: as much information as possible processed correctly.

Punkt!
Quite right and maybe the forums (all of them actually) should be split into those that display the final image by print and those that like to look at 100% on a monitor.

Different beast all together. The Epson printers are based on a 1440 weave pattern so any submultiple of 1440 will work better since it will require one less resampling. For B&W I use Quadtone Rip (QTR) and have found the best setting for at 360 or 720 DPI and a 1440 weave pattern in the RIP.

If I print color - which is only when my wife nags me enough - I use the Epson driver and a 360 DPI.
Sounds like you two gentlemen don't factor the human visual system and its acuity into your needs?
Can't speak for Laurence, however, in my case my workflow is judged by what is best hanging on the wall.
I understand, Truman.

Mine was a tongue-in-cheek comment to be honest. I should have put a smiley with it.

So, may I ask the relevant questions - what size is the picture hanging on the wall and from how far away do you view it?
 
On an Epson printer, an image looks great at 180 dpi, better at 360 dpi, even better at 720 dpi, etc.

The more information I can provide to a fixed-dimension output space (a print as opposed to a monitor), the better. Anyone who claims otherwise is just settling for "good enough."

The print is all about information: as much information as possible processed correctly.

Punkt!
Quite right and maybe the forums (all of them actually) should be split into those that display the final image by print and those that like to look at 100% on a monitor.

Different beast all together. The Epson printers are based on a 1440 weave pattern so any submultiple of 1440 will work better since it will require one less resampling. For B&W I use Quadtone Rip (QTR) and have found the best setting for at 360 or 720 DPI and a 1440 weave pattern in the RIP.

If I print color - which is only when my wife nags me enough - I use the Epson driver and a 360 DPI.
Sounds like you two gentlemen don't factor the human visual system and its acuity into your needs?
Can't speak for Laurence, however, in my case my workflow is judged by what is best hanging on the wall.
I understand, Truman.

Mine was a tongue-in-cheek comment to be honest. I should have put a smiley with it.

So, may I ask the relevant questions - what size is the picture hanging on the wall and from how far away do you view it?
 
I totally agree with Scott -

Chances are that 4320p will reach mass-market by the time SPP and such a "FF" sensor would mature, and if for Bayer and derivatives I can understand that pixel blur incurred by non-multiples might be beneficial, it seems to me that the Sigma sensor(up to Merrill) serves the 1:1 rendering better. As viewing my 2160p 50" computer screen most of the time from 2ft away, I can clearly see the pixels (100ppi), like on a 17" screen at 1440p from the same distance. As soon as 4320p connectivity matures and garden-variety GPUs are ready for 8k, I know what my next move will be. It does not seem obvious that all the unlit fiber will scale with the 360EB global traffic expected in 2018, so I take that their will be a pause for a while ; it took 33 years for NHK to go from 1125 to Super Hi-Vision in 2005.

24 x 36 sensor size is based on legacy (FF might as well have gone with the 8/10 aspect ratio) and besides transparencies, I don't know that printing paper will afford a dynamic range warranting sunglasses. If the size of the resulting photosite is optimal, for instance 38.4 mm wide sensor for a 5µm pitch like the Merrill, affording 7680 pixels across, why not ?

Chris
 

Keyboard shortcuts

Back
Top