first test of my Sigma 105mm F2.8 DG DN Macro Art

[Lou Jost]

impressive indeed!

so far I always was sceptical of sonys implementation of pixelshift because on most sample images I've seen there were zipper artefacts, but your example looks really clean. maybe people simply didn't use a sturdy tripod?
I wonder how it compares in resolution to my Panasonic S1R...

if you feel comfortable to send me your wafer to berlin I could do a comparison frame with a 8 pixelshift image on a S1R and a printing-nikkor. I would cover shipping costs of course.

Yes, that's the interesting question. I love my S1R and the 8x pixel shift photos are stunning. Those zipper artifacts are indeed just due to movement, as far as I can tell. This can include atmospheric turbulence (in landscape photos), where it can be irregular and patchy in the image, so it might not initially seem to be caused by motion.

[Beatsy]

Definite improvement! Lots of resolution in that lens. Pretty sure my "less than wow" results on the microscope (going from 4-shift to 16) are simply because the objectives were approaching or at diffraction limits. It was kind of a relief to see that as 16-shot pixel-shift workflow is "data heavy" and quite slow with FF. Worth it for some things, but not as a routine method IMO.

One thing I read about the "stair step" issue was that it mainly (only?) affected diagonal lines. The features on your chip are all aligned with the shift directions. If you're not too tired of taking test shots - the wafer set at 45 degrees might show it with a 4-shot to 16-shot comparison.

[chris_ma]

It was kind of a relief to see that as 16-shot pixel-shift workflow is "data heavy" and quite slow with FF.

yeah, one of the nice things with the S1R is that it assembles the images to a single raw file in camera. still kinda slow but otherwise a normal workflow.

can you batch process a stack of pixelshift images in the sony software?
and how many MB is the final RAW file?

[Beatsy]

...
can you batch process a stack of pixelshift images in the sony software?
and how many MB is the final RAW file?

Yes, you can leave it to process multiple files, any number. Each output file from 16-bit pixel shift, full frame, is approximately 2GB!

[chris_ma]

Yes, you can leave it to process multiple files, any number. Each output file from 16-bit pixel shift, full frame, is approximately 2GB!

wow, quite a lot of data!
strange thing is that an uncompressed 16bit TIF file of the same size is only about 1.4GB, so either they save it in 23bit (unlikely) or use a rather inefficient way of storing the data.

the PS RAW files of a S1R are about 350MB (edited typo). a 16bit TIF come in at 1.1GB, I've read that they use a very mild lossy compression.

[lothman]

One thing I read about the "stair step" issue was that it mainly (only?) affected diagonal lines. The features on your chip are all aligned with the shift directions. If you're not too tired of taking test shots - the wafer set at 45 degrees might show it with a 4-shot to 16-shot comparison.

I can give this a try over the next days. But what are we talking about? The Sigma lens or the implementation of pixelshift from Sony? For the latter we should this in a new thread.
regards
Lothar

P.S.: waver pieces mounted on a microscope slide are available, similar like this

[Lou Jost]

"the PS RAW files of a S1R are about 350GB."

That should be Mb!!

[chris_ma]

"the PS RAW files of a S1R are about 350GB."
That should be Mb!!

ah, good catch, I was getting carried away there for a second
I'll post a comparison to the S1R when the waver sample from Lothar arrives.

[Lou Jost]

"I'll post a comparison to the S1R when the waver sample from Lothar arrives."

I look forward to seeing that.

[lothman]

I look forward to seeing that.

Sony rulez (<- just kidding)

all shots from the centre at f2.8

so this is a crop from 4x-pixelshift taken at 45°:

the crop from above scaled to the 16x-pixelshift size:

and the same spot as 100% crop from the 16x-pixelshift:

please take in mind that direction of my two LED panes is from left and right. So shadow throw might have changed because I rotated the waver, but not together with the lightning. But IMO the gain in resolution is the same at 45° structures.

[lothman]

and some coins:

overview (about 1:1, f4)

single shot at 100% (slightly sharpened)

scaled up

and from 16x pixelshift (slightly sharpened)

[Chris S.]

Very well-done, interesting, and convincing tests, Lothar. The difference in that Lincoln penny is spectacular. Also, it was notable how well the diagonal lines held up on your rotated chip.

Sony rulez (<- just kidding)

As a Nikon shooter, this thread has me experiencing some Sony envy. Nikon does sensor shifting for vibration reduction, so one wonders why Nikon won't get off its butt and use the hardware apparently already in the camera to offer pixel shift for increased resolution.

And those Sigma Art lenses--I checked a couple of times to make sure you were shooting wide open, as I understood. Wow.

Then there are Sony lenses in focal lengths ignored by others--the 20-70mm and 12-24mm, for example. I could totally use these for landscape and travel.

--Chris S.

[FotoChris]

@Chris S.

Pixel shift is great for such tests; for 1:1 at f4 you should be able to get something like 70MP of useful data from the sensor, maybe slightly more in the center portion. Anything beyond that my be difficult to achieve.

For practical purposes... well let's say it's very difficult to get it to work on a daily basis, sometimes it works well under very, very special circumstances. But most users won't have the patience, necessity or...well..the correct 'subject' in order to make it work - and if it doesn't work people like to blame the manufacturer because people would think it should "just work". I think that's why many manufacturers tend not to integrate it into their cameras, I still remember the reactions to the initial issues when the Fuji GFX100(S) pixel shift was first introduces to the masses.
People were...less than impressed.

And even when the conditions are (theoretically) perfect - like in the last coin image - artefacts like "scan lines" are very common and very likely to happen. Whether it's due to slight vibrations, minute movements, changes in light,... and all that is even more critical for macro shots - and then you may not notice until it's too late and waste a lot of time in re-creating the image or fixing it in post.
It reminds me of those large format and medium format scan backs / scan cameras like the Seitz 617 or Better Light.
I mean if everything goes right and the starts align and all the gods in the universe give their blessing - then pixel shift can be useful for still life. But I would consider that to be the exception.

Generally it's easier to go for a higher magnification and stitch images.

So I understand why Nikon doesn't feel like they have to rush into pixel shift on their cameras.

[Lou Jost]

@Chris S.

Pixel shift is great for such tests; for 1:1 at f4 you should be able to get something like 70MP of useful data from the sensor, maybe slightly more in the center portion. Anything beyond that my be difficult to achieve.

For practical purposes... well let's say it's very difficult to get it to work on a daily basis, sometimes it works well under very, very special circumstances. But most users won't have the patience, necessity or...well..the correct 'subject' in order to make it work - and if it doesn't work people like to blame the manufacturer because people would think it should "just work". I think that's why many manufacturers tend not to integrate it into their cameras, I still remember the reactions to the initial issues when the Fuji GFX100(S) pixel shift was first introduces to the masses.
People were...less than impressed.

And even when the conditions are (theoretically) perfect - like in the last coin image - artefacts like "scan lines" are very common and very likely to happen. Whether it's due to slight vibrations, minute movements, changes in light,... and all that is even more critical for macro shots - and then you may not notice until it's too late and waste a lot of time in re-creating the image or fixing it in post.
It reminds me of those large format and medium format scan backs / scan cameras like the Seitz 617 or Better Light.
I mean if everything goes right and the starts align and all the gods in the universe give their blessing - then pixel shift can be useful for still life. But I would consider that to be the exception.

Generally it's easier to go for a higher magnification and stitch images.

So I understand why Nikon doesn't feel like they have to rush into pixel shift on their cameras.

I have to disagree with this, based on my experience with Olympus and Panasonic implementations. Yes, it is hard in the field. But for studio work it is quite reliable and MUCH easier than stitching.

[FotoChris]

I have to disagree with this, based on my experience with Olympus and Panasonic implementations. Yes, it is hard in the field. But for studio work it is quite reliable and MUCH easier than stitching.

What I mean is that the actual need and actual benefits for pixel shift are rather small compared to the additional work, try/error and fixing stuff.
If you need like a few pixels more then stitching two single images together is more reasonable (to me) than setting everything up for a 4 or 16 pixel shift because the useful scenarios are so limiting.

The most important question is: How big do you print? How much do you crop? How many pixels are necessary in real life?

It's not "just" about it being "quite reliable" in a highly controlled studio environment, it's about being pragmatic and useful enough for most people as to be implemented by a company like Nikon.

They may yet chose to implement it, it took a lot of companies long enough to integrate focus stacking (or rather focus stepping) but that has a wide range of use for a lot of photographers. Quadrupling the resolution in very restrictive situations is not.