Fujifilm is working on a 400MP pixel shift mode for the GFX100

[BernardLanguillier]

Very happy to see this coming!

Cheers,
Bernard

[Avalanche]

I have read quite a few threads about that new pixel shift feature. I do see that there are use cases, but most of them come with strong limitations or are targeted to a very specific group of photographers. I am wondering if those software development resources would not have been better spent on other improvements that target a larger audiences. On the other side that feature might just be a side product of the pixel shift implementation of the new X-T4 (under the assumption that it will have a similar IBIS setup to the GFX 100 and that they are planning on offering that feature for other cameras as well).

Let's see how their implementation turns out. This is not a choice for customers after all.

[JaapD]

I don’t expect pixel shift functionality with the upcoming X-T4. This would be very difficult to implement on a 6x6 X-Trans color filter array.

Regards,
Jaap.

[BernardLanguillier]

I have read quite a few threads about that new pixel shift feature. I do see that there are use cases, but most of them come with strong limitations or are targeted to a very specific group of photographers. I am wondering if those software development resources would not have been better spent on other improvements that target a larger audiences. On the other side that feature might just be a side product of the pixel shift implementation of the new X-T4 (under the assumption that it will have a similar IBIS setup to the GFX 100 and that they are planning on offering that feature for other cameras as well).

Let's see how their implementation turns out. This is not a choice for customers after all.

At least this pretty much puts an end to the life of the Hasselblad H system... not that I am happy about that as a H6D-100c owner, but the only specificity that could still have convinced new photographers to buy into the system at this point in time is now gone and the revenue stream associated to it as well.

Some rich Chinese may still prefer the shades of grey of the H body over the dark black of the Phaseone XF... but that's pretty much it.

Cheers,
Bernard

[Paul2660]

Fuji has avoided pixel shift on x-trans cameras. They did not pursue it with the X-H1 which has IBIS.

Odds are the XT4 will not have it either due to same issue.

GFX100 has traditionally designed Bayer pattern which allows the use of pixel shift design considerations more acceptable.

Paul C

[Jim Kasson]

Partly in response to this thread, I've done a theoretical study on the effect of pixel shift on resolution:

https://blog.kasson.com/the-last-word/does-pixel-shift-increase-resolution/

Jim

[BJL]

Thanks Jim!
I can vaguely understand why the MTF is unchanged with the “every pixel sampling” of your luminosity-only model. What is less clear is the effect with Bayer CFA. In the simplest case of four shifts by full pixels, one measures red, blue and green (twice!) at each location, so like Foveon X3 and a bit more (Dare I call it X4?). That seems good for something, changing or eliminating the spatial averaging needed in Bayer demosaicing. Reports are that the shift modes of Olympus and Pentax do add some visible benefit.

But with luminosity resolution more important for our eyes, it might not add much in practice.

[Jim Kasson]

Thanks Jim!
I can vaguely understand why the MTF is unchanged with the “every pixel sampling” of your luminosity-only model. What is less clear is the effect with Bayer CFA. In the simplest case of four shifts by full pixels, one measures red, blue and green (twice!) at each location, so like Foveon X3 and a bit more (Dare I call it X4?). That seems good for something, changing or eliminating the spatial averaging needed in Bayer demosaicing. Reports are that the shift modes of Olympus and Pentax do add some visible benefit.

But with luminosity resolution more important for our eyes, it might not add much in practice.

The 4-shot sampling doesn't increase sharpness, but, done right, completely eliminates the false-color artifacts associated with the Bayer CFA. In addition, it avoids any loss in sharpness caused by interpolation during Bayer demosaicing.

The 16-shot sampling does all that and in addition allows the elimination of some more aliasing.

There are undeniable advantages (and disadvantages) to pixel-shift shooting, but pure MTF is not one of them, because the sampling window is 4 times are area of what it should be for a sensor with half the pitch of the actual sensor.

Jim

[BJL]

At least this pretty much puts an end to the life of the Hasselblad H system ... the only specificity that could still have convinced new photographers to buy into the system at this point in time is now gone and the revenue stream associated to it as well.

It could be so, unless the H system finally gets a new "full frame" back exceeding the detail (pixel count) of the new 44x33 mirrorless systems.  Because after the technical input in this thread, my prediction is only slightly changed: the only "specificity" that I see being strong enough to maintain a market for new gear in the far bulkier and more expensive 645 SLR-based cameras is larger sensors that give more detailed images. And even that is a thin advantage, with an only modestly larger format (47% area, 21% linear); the best that 645 SLRs can hope for is Sony to keep making 53x40mm sensors with the same unit cell as in 44x33mm sensors, giving a 21% higher (linear) resolution and so prints of 47% greater area at equal PPI.

Most other advantages typically claimed for larger formats seem mostly neutralized by things like improving multi-shot (DR stacking, etc.), leaving mainly "more detail" and the love of big, bright OVF images. The OVFs might provide enough reason for some people to keep using 645 SLR gear, but I doubt it will be enough to attract enough new customers and sales for the longer term.

[EricWHiss]

Partly in response to this thread, I've done a theoretical study on the effect of pixel shift on resolution:

https://blog.kasson.com/the-last-word/does-pixel-shift-increase-resolution/

Jim

Interesting, and thanks!

My experiences using the multi-shot and micro step on the Imacon/Hasselblad  and Sinar backs really made me a believer, but I've been a bit underwhelmed by the pixel shift on the newer cameras based on my expectations from the older backs.   I've just got the Sony A7rIV and haven't yet tried their 16 shot mode.   My first test with multishot, I taped up a bunch of stuff to a wall including a dollar bill and shot some images.  The things that really impressed me the most was not the increase in hard line things like the print and the cloth fibers in the dollar but stuff like my finger print on the scotch tape used to fix the bill to the wall.  This was subtle thing, very faint, and it wasn't there at all in single shot and appeared with remarkable clarity in the multi-shot.   And the multishot images just had so much more palpable illusion of depth to them when you stepped back and looked at the whole image and not just 100%.     

Each frame is taken with the same sensor regardless of single or multi -  so is it right to assume a smaller pitch in your calculations?  In a way with shift you're actually sampling 4x the area for each pixel and in my empirical experience this results in 1.5 - 2 stops more usable DR and being able to stop down further without noticing diffraction softening - exactly what you'd get from a larger sensel and opposite of what you'd get with a smaller sensel.   And also why assume a monochromatic sensor?  One of the biggest advantages with pixel shift/multi shot is that you're recording true color information for each site.  I think there must be a better way to evaluate the loss of information between an interpolated RGB data set and a true measured RGB data set - it's not so much that pixel shift has more resolution as it has more accurate data.     I think another question mark is how the frames are assembled since it's proprietary software and there are some differences in methodology.

[Jim Kasson]

Each frame is taken with the same sensor regardless of single or multi -  so is it right to assume a smaller pitch in your calculations?

Yes. The pitch of the reconstructed image is finer than the pitch of the sensor.

  In a way with shift you're actually sampling 4x the area for each pixel and in my empirical experience this results in 1.5 - 2 stops more usable DR...

Yes, pixel shift does increase DR, as would taking 16 images of the same subject without shifting.

...and being able to stop down further without noticing diffraction softening...

The measurable diffraction softening in the absence of aliasing is the same. However, you have the ability to sharpen the pixel-shift image more since it has fewer aliasing artifacts.

- exactly what you'd get from a larger sensel and opposite of what you'd get with a smaller sensel.   And also why assume a monochromatic sensor? 

To make the analysis tractable, and to get away from issues about the sharpness and aliasing changing with the demosaicing algorithm.

-One of the biggest advantages with pixel shift/multi shot is that you're recording true color information for each site. 

Absolutely! There is a dramatic reduction in color aliasing.

I think there must be a better way to evaluate the loss of information between an interpolated RGB data set and a true measured RGB data set - it's not so much that pixel shift has more resolution as it has more accurate data.

That last sentence is precisely my point!

    I think another question mark is how the frames are assembled since it's proprietary software and there are some differences in methodology.

Yes. A reason to assume a monochromatic sensor is to sidestep all that. It's complicated enough as it is.

Jim

[bbrantley]

So...  after all the math is over and done with (thank you Jim and others), I'll just add:  go have a look at the Hasselblad samples from five or six or however many years ago when they came out with their big campaign, if you haven't before.  The difference is incredible.  The single-shot images, which were fine a few seconds before, immediately look a bit like mud.

If you shoot static things, not on a timeline, as a way of making art, and your goal is to print large (and one day, "display large," of course), this is a huge step up, indeed.

I hope Phase can develop a stabilization-based body one day, not just so that they remain competitive, but also because then Doug will start telling us how awesome it is. 

[bbrantley]

I hope that they do Hi-res with an update. Many will smile...

Seriously!

[Doug Peterson]

So...  after all the math is over and done with (thank you Jim and others), I'll just add:  go have a look at the Hasselblad samples from five or six or however many years ago when they came out with their big campaign, if you haven't before.  The difference is incredible.  The single-shot images, which were fine a few seconds before, immediately look a bit like mud.

A major driver in that case is that Hasselblad has very dated/mediocre single shot raw processing in their Phocus software.

If someone wants to make a comparison they should do so between a best-in-class single shot and best-in-class multishot system.

[BJL]

A major driver in that case is that Hasselblad has very dated/mediocre single shot raw processing in their Phocus software.

If someone wants to make a comparison they should do so between a best-in-class single shot and best-in-class multishot system.

Multi-shot seems to show significant IQ gains with the Olympus and Pentax implementations too, so it seems hard to put it mostly down to poor handling of regular one-shot conversions.

[bbrantley]

A major driver in that case is that Hasselblad has very dated/mediocre single shot raw processing in their Phocus software.

That may be so, I wouldn't know.  I just downloaded the TIFFs from their site.

If someone wants to make a comparison they should do so between a best-in-class single shot and best-in-class multishot system.

That sounds like a great test.  As soon as Fuji finishes shipping this feature, we'll all have a chance to compare GFX100 multishot to GFX100 singleshot.

In addition, I suppose we could also compare GFX multishot to Phase singleshot. 

But, seriously:  I'm not shelling out 3-5x the money for a Phase system with what, for me, is substantially inferior shooting ergo.  So, the compelling comparison for me will be:  does multishot improve the results with the camera I can afford.  I am pretty sure I know what the answer will be already.  If it also happens that GFX multishot is better in some ways than Phase 150MP singleshot, that's nice to know in an academic sense.

[Jim Kasson]

That sounds like a great test.  As soon as Fuji finishes shipping this feature, we'll all have a chance to compare GFX100 multishot to GFX100 singleshot.

I see no reason why, at a pixel level, it won't look like the a7RIV pixel shift shots. Think of the a7rIV sensor as a crop of the GF 100 sensor.

Jim

[EricWHiss]

A major driver in that case is that Hasselblad has very dated/mediocre single shot raw processing in their Phocus software.

If someone wants to make a comparison they should do so between a best-in-class single shot and best-in-class multishot system.
[/quote

I've done that, still impressed with the CF 528 which is Hasselblad - no problems with the software.    Workflow is simpler with single shot, but I won't be saying it's best even with the 150mp tri color.  TPO really missed an opportunity not going with the multi-shot.   If you are doing art reproduction subtle stuff gets missed without having the non-interpolated color.  Very curious to see how the Fuji 100mp camera does with pixel shift.    Really seems to me that Phase missed an opportunity to be as you write 'best in class" (what a tired expression) by not offering a multi-shot option.   

[bbrantley]

I see no reason why, at a pixel level, it won't look like the a7RIV pixel shift shots. Think of the a7rIV sensor as a crop of the GF 100 sensor.

Jim

Well, to Doug's point, if there are substantial differences in the magic software soup, then things could turn out differently.  This isn't (necessarily) just about hardware equivalency.

If you made me guess, though:  sure.  It seems reasonable to suspect that it will be about the same.

[Doug Peterson]

A major driver in that case is that Hasselblad has very dated/mediocre single shot raw processing in their Phocus software.

If someone wants to make a comparison they should do so between a best-in-class single shot and best-in-class multishot system.

I've done that, still impressed with the CF 528 which is Hasselblad - no problems with the software.    Workflow is simpler with single shot, but I won't be saying it's best even with the 150mp tri color.  TPO really missed an opportunity not going with the multi-shot.   If you are doing art reproduction subtle stuff gets missed without having the non-interpolated color. 

That's not the evaluation of the overwhelming majority of high-end cultural heritage institutions (museums, libraries, archives). Very very few of them are still using multi-shot cameras, in contrast to 15-20 years ago when many (most?) of them did. Specifically, many (most?) of them are now our clients.

As I've said numerous times in the past: there are use cases where modern multishot can still be helpful, but they are rare, the difference is modest, and the constraints are significant. Most of what you see written about multishot comes from the golden era of multishot (when 16mp to 22mp sensors dominated and poor single-shot raw processing).

If Fuji adds a multishot mode via firmware the users should try it; free features added after launch are an unequivocal good thing – on that, Fuji and Phase One agree. But if those users think firmware will magically turn their system into the equivalent of a 400mp camera, as many will based on how multishot is advertised, there will be a lot of disappointment.