[photoptimist]

The equating of a 36MPix pixelshift image to a 144 MPix Bayer filter image isn't totally wrong.

A 144 MPix Bayer filter RAW image only has 36 MPix of red channel data and 36 MPix of red channel data. Both of these channels have to upsampled (that's what demosaicing does) to fill in all the missing data in red and blue. A 144 MPix Bayer filter RAW image only has 72 MPix of green data that is also upsampled.

36MPix pixelshift image upsampled to 144 MPix would have the same resolution in red and blue as a 144 MPix single shot Bayer filter image but the upsampled pixelshift image would be a bit softer in the green channel than the 144 MPix camera image. The 144 MPix single shot Bayer filter image might suffer from some aliasing artifacts while the pixelshift image might suffer from subject movement or light nonuniformity artifacts.

That's why I said "it's complicated"

For subject matter that is totally motionless and lit with a stable light source (not an inconsequential prerequisite), pixelshift image will compare favorably with Bayer filter images of nearly double the sensor pixel count.

[surfar]

the claim the pixel shit k-1 was a “144mp” image. I didn’t think it was accurate.

Yes, I threw it in there because that was "the talk" back in 2016.

Now, Olympus on the other hand with "Hi RES" can put out 50mp and 80mp images from their 20mp sensor.

[biz-engineer]

For subject matter that is totally motionless and lit with a stable light source (not an inconsequential prerequisite), pixelshift image will compare favorably with Bayer filter images of nearly double the sensor pixel count.

Yes, but with a larger format camera having a higher pixel density, you may need to focus-stack due to DoF, unless you have a TS lens. If you need to focus stack, that's a multi-exposure, and there come the problem of subject motion as well. Based on that, my interest in pixel shift was suddenly revived because I came up with a new idea about the nature of various subjects, and how I could use camera settings to fit a quick pixel shift exposure within a stack of normal exposures.

[fs999]

From Dpreview...

Attached Images

 

[biz-engineer]

Pretty close. ~4K crop of output files prepared for A1 size prints.
GFX100_A1_PRT_SAI-Softness_cr.jpg - Google Drive
MK2_0149_A1_PRT_SAI-Softness_cr.jpg - Google Drive

[surfar]

geek post warning

OK, Biz...how about comparing camera bodies that cost a similar amount with each other?

But not immediately because the New Fuji isnt available for a while.

What would be definitely interesting is OM1 HiRes...Fuji crop 40mp Pixel shift...Pentax K1 Pixel shift.All these cameras cost around the same but are all different sensor sizes.

---------- Post added 09-13-22 at 09:43 PM ----------

Pretty close.

I zoomed in and they are not close.Colours in the Pentax where there shouldn't be any and no where near the same resolution.

[biz-engineer]

I zoomed in and they are not close.Colours in the Pentax where there shouldn't be any and no where near the same resolution.

You should zoom in 50% on a 4K monitor or 25% on an HD monitor. If you zoom in 100%, you are looking at a print 2x or 4x the A1 size, which is not the print size I'm looking at. If I'd print 2 meters wide, of course the difference between K1 and GFX100 would be more obvious (not not night & day either).
If you zoom 50% or 25% and you can see a meaningful difference (beside the obvious moiré patterns), then you must have very very good eyes.

---------- Post added 13-09-22 at 15:25 ----------

But not immediately because the New Fuji isnt available for a while.

I've downloaded Fuji XH2 40Mp files and errrr, my reaction was mehhh, not impressed compared to K1, not bad, but rather soft or a bit washed out, or over processed, a bit of a stretch, I don't know...something just doesn't look right. I much prefer the look of K3III images. But yeah, I recognized the 40Mp images are impressive for apsc format.

---------- Post added 13-09-22 at 15:09 ----------

From Dpreview...

I don't use the comparison from DPReview tool, because the tool zoom at 100%. The print comparison of DPR tool downsizes to 8Mpixels, which obviously is outdated.

---

Last edited by biz-engineer; 09-13-2022 at 06:26 AM.

[photoptimist]

Yes, but with a larger format camera having a higher pixel density, you may need to focus-stack due to DoF, unless you have a TS lens. If you need to focus stack, that's a multi-exposure, and there come the problem of subject motion as well. Based on that, my interest in pixel shift was suddenly revived because I came up with a new idea about the nature of various subjects, and how I could use camera settings to fit a quick pixel shift exposure within a stack of normal exposures.

Hmmmm.. It might be a bit trickier than it seems.

If you are comparing K-1 Pixel shift to the GFX100 by pixel peeping, then the acceptable circle of confusion and DoF depends on the properties of the sensor:

In the red and blue channels of a Bayer sensor image, the acceptable circle of confusion would be tied to twice the pixel spacing of the sensor. In the green channel, the acceptable circle of confusion would be tied to square-root-of-two (1.414) times the pixel spacing of the sensor.

For a pixel shift image, the acceptable circle of confusion would be tied to one times the pixel spacing of the sensor.

If you are comparing K-1 Pixel shift to the GFX100 by making prints of a certain modest size, then the circle of confusion and DoF depend on human visual acuity and how the print is viewed. That circle of confusion on the print then relates back to a circular of confusion on the sensor through the ratio of the sensor format size to the print size.

[jeffdrew]

Isn’t the k1 pixel shift only 36mp? Albeit with high color accuracy.

As a geezer, I’ve lost the need for “size” or studio sized kits & projects. A K1 is more than a handful for me, as are the bigger Fuji kits. Pentax K3ii & 3 are more my speed, and loving all the new post processing tech. Even my Fuji x10 is back in my bag. Uncle, I respect your observations and extend best wishes! Thx!

[UncleVanya]

As a geezer, I’ve lost the need for “size” or studio sized kits & projects. A K1 is more than a handful for me, as are the bigger Fuji kits. Pentax K3ii & 3 are more my speed, and loving all the new post processing tech. Even my Fuji x10 is back in my bag. Uncle, I respect your observations and extend best wishes! Thx!

Lol. Only should be in quotes. 36mp is plenty. I shoot 24mp apsc and 42mp ff and I wouldn’t care if the ff was 36mp. The camera makes good records of what the photographer tells it to - that’s all that’s needed for me.

[Rondec]

It is an interesting comparison and mainly says to me that current cameras are quite a bit more powerful than what most of us need, considering our normal print/viewing sizes. I understand that if you look at these photos at 100 percent you can see a definite difference, but that isn't the typical way in which these will be viewed.

I love pixel shift -- mainly because it gives me a lot more latitude when processing an image than a standard K-1 image allows for. The problem of course is that pixel shift requires a relatively still scene for there to be benefit. The GFX-100 should produce similar results from a single shot -- although depth of field could be an issue if you can't stop down enough.

(Requisite K-1 pixel shift image)

[biz-engineer]

For a pixel shift image, the acceptable circle of confusion would be tied to one times the pixel spacing of the sensor.

Practically, except if the subject is distant, or if the subject is flat parallel to the camera sensor, depth of field each side of the focus plane is taxing sharpness much more quickly than anything else. Shooting on tripod becomes key to success.

The GFX-100 should produce similar results from a single shot -- although depth of field could be an issue if you can't stop down enough.

The answer to that is to shoot a focus stack and blend them in post. For long exposure photography, stacking has the advantage over stitching, because it is much easier to line-up a stack of exposures than it is to line-up a wide angle long exposure with a stitched panorama.

[fs999]

I don't use the comparison from DPReview tool, because the tool zoom at 100%. The print comparison of DPR tool downsizes to 8Mpixels, which obviously is outdated.

But you can download both raw files !

[Rondec]

Practically, except if the subject is distant, or if the subject is flat parallel to the camera sensor, depth of field each side of the focus plane is taxing sharpness much more quickly than anything else. Shooting on tripod becomes key to success.


The answer to that is to shoot a focus stack and blend them in post. For long exposure photography, stacking has the advantage over stitching, because it is much easier to line-up a stack of exposures than it is to line-up a wide angle long exposure with a stitched panorama.

Sure you can do focus stacking, but then you have a second step, the same as with pixel shift. My understanding with focus stacking is that it too can have issues if there is a lot of movement in a scene, beyond which you have to have a decent computer to process four or five 100 megapixel RAW images. Obviously people do it and get good results, but the limitations are actually not too different from pixel shift.

[biz-engineer]

I zoomed in and they are not close.Colours in the Pentax where there shouldn't be any and no where near the same resolution.

After looking again in DPR studio comparison with pixel shift enabled, I realized there is a problem with Silkypix handling pixel shift files. So, I converted the K1 pixel shift image with Rawtherapee, and I converted the GFX100 image again because there was a miss-match of white balance. Uploaded both crops again.
GFX100_SPD_PRT_cr_SAI-Softness.jpg - Google Drive
MK2_0149_RT_PRT_cr_SAI-Softness.jpg - Google Drive

My understanding with focus stacking is that it too can have issues if there is a lot of movement in a scene

Correct. Movement can be used to intentionally convey motion in images, via long exposures.
One way to achieve high resolution is to stitch images, another way to get higher definition images is to use pixel shift.

Stitching:
Stitching is a proven technique to achieve high resolution, although it can be problematic for long exposure photography and/or wide angle.
For example, stitching 5 vertical exposures of 2 minutes each is problematic when the light is changing quickly. After 10 minutes the sunset is gone and the color temperature of the last exposure is completely different from the color temperature of the first shot.
One solution to deal with long exposure stitching is to use a zoon lens, capture one frame of the scene with long exposure e.g 2 minutes, then zoom in an capture a sequence of images for the stitch, then line up the master frame with the stitched images.
But the big issue with lining up a master frame with a stitched panorama is that it give headaches to the stitching software and it can distort the master frame to fit exactly the panorama. That's a problem pixel shift doesn't have.

Pixel shift:
So my idea is to combine a quick pixel shift exposure with long exposures, and stack them into a composite image.
Now , the problem is that pixel shift is also sensitive to change of light during exposure, but I work around that.
The work around is to use an ND filter and open up the lens diaphragm so that the pixel shift exposure time doesn't exceed a few seconds.
Then stop down the lens to f22 or more if possible, so that the exposure time gets much longer, and reach a 2 minutes total exposure time via a composite average or interval average.
Then overlay the pixel shift frame with long exposure frames and use a mask to combine smooth and sharp details areas in one image.
I tried to do a pixel shift without ND filter, then add the ND filter for the long exposures, but that's not working very well because the color cast of the filter create a color mismatch between the long exposure composite and the pixel shift image.
When the light is changing quickly, e.g at sunset/dawn/dusk,the even better workflow that I haven't tried yet if to sandwich the pixel shift exposure between two long exposures, so that the average light of the long exposures composite is similar to the light level of the pixel shift exposure.