Pentax K3 II: SuperResolution Pixel Shift Mode vs StdRes: Tarnished Silver Dragonfly Pitcher
Related: Adobe, Adobe Camera Raw, Adobe Camera Raw Super Resolution, digital sensor, Electronic First Curtain Shutter, Electronic Shutter, exposure, James K, noise, Pentax, Pentax APS_C, Pentax DSLR, Pentax K-3, Pentax K-3 II, Peter H, pixel shift, shutter, true-color sensor or image
See my Pentax K1 wish list at B&H Photo.
See also my in-depth review of the Sigma 35/1.4 DG HSM Art, used for this study.
Separately, I've added new comments on the SuperResolution: Checkerboard Effect page discussing concerns about general applicability of the pixel shift feature.
This evaluation pits the Pentax K3 II SuperResolution pixel shift mode against its standard conventional resolution Bayer recording. SuperRes mode in discussed in more detail in the review of the Pentax K1.
The subject here is a tarnished silver pitcher with extensive patina, specular reflections, and some out of focus background metal areas that are darker and smoother. Very fine details including reflections, speckled tarnish and scratches yield a very good idea of the performance differences.
Pentax K3 II SuperResolution: Tarnished Silver Pitcher
It’s not just about detail: there are implications for dynamic range and noise and 3D feel as well.
This and the previous examples with the Pentax K3 II are a particularly persuasive argument that the full frame Pentax K1 could become a powerful specialist tool for true-to-life imagery at a rock-bottom price (relatively speaking, at about $1000 less than the Nikon D810). The 36-megapixel full-frame Pentax K1 pixels are 55% larger in area than the APS-C 24-megapixel K3 II pixels, thus more forgiving of lens performance and promising even higher quality.
Picture of a pitcher. Say that fast three times!
PENTAX K-3 II + lens @ 35mm
[low-res image for bot]
Peter H writes:
I'm starting to agree with you Lloyd. I think the examples you are showing of the K3 II's SuperRes mode are the highest quality images I have ever looked at (and yes, I do have a high quality calibrated NEC monitor).
Might some of the difference in exposure between StdRes and SuperRes modes on the silver pitcher be a result of the differences between manual shutter and EFC? I recall Michael Reichmann (Luminous Landscape) commenting recently that he thought the technology behind electronic first curtain wasn't quite there yet (he said it was probably a couple of years away before it would be 100%). Apparently EFC can be prone to slight exposure errors (and he wasn't talking about flash situations).
Also, what exposure time penalty is there in SuperRes mode? How long does the camera take to make its 4 exposures, vs a normal exposure?
DIGLLOYD: The K1 is an exciting prospect with its 55% larger pixels, and if there is any slop on the pixel shifting, there should be more margin for error on the K1. So I deem the K3 II exciting for what it suggests for the K1.
Exposure: each pixel gets 4 exposures (recording R + G + G + B) and somehow those have to be combined. Assessing noise in individual channels shows that in the finished RGB image, noise in the red and blue channels is substantially lower than with the StdRes image, and yet the R and B values are single exposures only (two G values at each pixel). In other words, the rendering to RGB incorporates benefits into all channels. So it would not be surprising to see deviations of some kind, rather than presume a shutter explanation. Which is not to state that the EFC shutter is uninvolved; it could be.
Exposures: 4X as long plus ~2 seconds "processing image". It’s not much of a downside in practice, because SuperRes mode demands a bare minimum 2 second self timer delay and absolutely no movement of the camera. So SuperRes mode is not a casual affair and I have my concerns about the viability of SuperRes pixel shift mode for field use.
James K writes:
Sinar backs are the pixel shift choice of many ultra detail hunters. A Sinar P3 and a Sinar H86 back and away you go and away goes around $35,000.
DIGLLOYD: exactly. The Pentax K1 is about $1799. With any luck, Sigma will offer all its Sigma f/1.4 DG HSM Art lenses for the K1. If the need strikes you, the Hasselblad H5D-200c Multi-Shot Medium Format DSLR Camera Body is only $45,000, and for that price you get a crummy low-res LCD and all the lenses you want, so long as you pay for those too.
The K3 II image quality suggests to me that excepting the super exotic high end, the K1 may be an outrageous value for any studio photographer.
Kevin S writes:
The three dimensionality & micro contrast of the SRPSM reminds me of the performance of the Otus lenses.
Could we all collectively put in a request to Canon to come up with a similar strategy?!? Can you imagine the resolution & micro contrast with Otus Lenses & SRPSM? So Santa, we now want Otus lenses on a Canon 100 megapixel camera that has Super Resolution Pixel Shift Mode. By November please. :-)
DIGLLOYD: Gee I hadn’t thought of that. :) I’d like to see this technology on Sony (as noted last year) plus Nikon and Canon, as it would move the quality level up more than any development in years. Sony already has IBIS, so why the option is absent is unclear.
Karsten W writes:
I am quite impressed with the pixel-shift samples from the Pentax K3 II, and I am really looking forward to your full review of the K-1!
Here’s a question that I have in the meantime: how much advantage do you gain by the exact pixel-by-pixel shifting as implemented by Pentax compared to merging multiple randomly shifted exposures (for example with PhotoAcute). A while ago, I did a trial with my Nikon D800E on a tripod (and a rail), and I carefully shifted between multiple exposures to the left or right for fractions of a millimeter. I then combined 4 or 8 exposures with PhotoAcute and got a rather nice improvement of the resolution.
I assume I did not gain as much as with the accurate pixel shift (because the colors cannot be restored as accurately?) but it did help quite a bit with some of the details. This seems to be a poor man’s implementation of pixel shifting and could enhance the quality of certain images (landscape, still lifes etc). Also, it can be done fast (for example with a macro rail and a pano head so that the camera can be moved rapidly left or right by very small increments to avoid parallax problems).
Any comments on this? Unfortunately, I do not have access to a Pentax K3 but maybe you are willing to perform a real test, sensor shift versus manual shift?
Thanks and keep up the good work!
DIGLLOYD: Multiple randomly shifted exposures can also suffer from slight movement/skew (even a few microns is an issue). Done right, similar effects can be had, but there is no guarantee as to how things overlap, and shooting 4 or 8 exposures manually takes more time (a lot more time given the delays in the EFC mode with the D810), and time itself is a problem (changing light color or intensity). And in the end the results are subject to chance. Personally I’m not a fan of the tedium involved, but it may appeal to some. The use of special software like PhotoAcute might avoid the checkerboarding issues to some extent, but that too forces one out of a standard workflow (ACR in Photoshop for me).
Bottome line as I see it: the K-1 may may a reliable special purposes tool, such as for studio work with its pixel shift feature. Everything else, I’m sticking with Sony and Nikon and Canon, particularly outdoors where camera stillness cannot be guaranteed across exposures (wind in particular). Plus the lenses I want to use just aren’t there for Pentax (Otus in particular, but many others also).
