Pixel shift vs no shift example

[Lou Jost]

This is an extension of Beatsy's recent pixel shift post. This is a Sunset Moth at m=2.67 on a FF S1R. First the whole image (an incomplete stack, some the hairs were outside of the stacking range):

Next is one of the images that went into this stack, in a pixel-shifted version and a normal version, enlarged to 200% and 400 percent respectively:

At left is the eight-shot pixel-shifted image, and at right is one of the eight shots that helped create the shifted image. At that magnification, the ribs of the scales are just barely resolved in the normal-sized image, and only if they more or less align by chance with the pixel grid of the sensor. The shifted image is able to resolve all the ribs that are in focus (remember this is a single unstacked image of a very three-dimensional object). No sharpening or adjustments, and default sharpening in ACR is turned off.

[seta666]

Thanks for posting this, that is a big improvement in resolution, and I guess in DR too. Is the single shot the first one of the 8 shoot series? maybe as the sensor is moving there is some loss in sharpness in the following ones.

How long does it take each 8 shoot PS vs a normal shot?

[chris_ma]

Is the single shot the first one of the 8 shoot series? maybe as the sensor is moving there is some loss in sharpness in the following ones.

in my experience all the sigle frames have the same sharpness, but I never tested higher then 2x magnification.

How long does it take each 8 shoot PS vs a normal shot?

shooting tethered with lumix tether into a lightroom watch folder, I get:

- 2sec delay
- less then 1sec for the 8 exposures
- about 10 secs for the camera to assemble the pixel shift image
- about 10 secs for the file to be transferred to the computer and show up in lightroom
(btw make sure you use a USB cable that supports true 5gbps or higher, the first I used took 7 seconds longer)

I could imagine that shooting to a fast memeory card would be slightly faster (I didn't try since I need tethered shooting for my workflow), so with reducing the delay and optimising things one could probably get under 20sec for a pixel shift image.

single frame is pretty much instant and the transfer takes a couple of seconds.
chris

[seta666]

in my experience all the sigle frames have the same sharpness, but I never tested higher then 2x magnification.
-------------------
................so with reducing the delay and optimising things one could probably get under 20sec for a pixel shift image.

I would be interested in seeing a comparison between first and last shot at 40X, maybe those fast movements could have some detrimental effects

20s seems like a lot, maybe not something I would use for very deep stacks

[Macrero]

The perceptible improvement in resolution/fine detail depends on the subject. With scales, wafers, etc, it is pretty clear when viewing at a reasonable size/resolution. But in my experience, when shooting insects, in most cases the improvement is subtle and (for me) not worth the time and effort required to stack with Pixel-shift, especially if the final destination of the photo is the web.

To each his own, I guess, but I recently sold my S1R and I am currently using a humble X-T3. I don't regret it at all.

Looking forward to seeing some great pixel-shift stacks though, so far I've seen a lot of comparisons and demonstrations about how great pixel-shift is, yet very few (if any) actual piixel-shift stacked photos at a resolution that allows to appreciate the benefits.

[chris_ma]

20s seems like a lot, maybe not something I would use for very deep stacks

the pixel shifted files are also around 350MB per RAW file and 1GB per 16bit tif frame, so a 500 frame stack not only would take 3 hours to capture, it would also create 170GB of RAW and 500GB of intermediate data.

I agree that for web use and similar it's definitely not worth it, there are some technical applications where there can be a real benefit though (and they are just fun to look at

[seta666]

20s seems like a lot, maybe not something I would use for very deep stacks

the pixel shifted files are also around 350MB per RAW file and 1GB per 16bit tif frame, so a 500 frame stack not only would take 3 hours to capture, it would also create 170GB of RAW and 500GB of intermediate data.

I agree that for web use and similar it's definitely not worth it, there are some technical applications where there can be a real benefit though (and they are just fun to look at

The 40mpx PS from a E-M5 mkII would be more than enough for me and I would probably export those 40mpx RAWs as 24/32mpx TIFF os something like that. I made 90x60cm prints for an expo with a NEX-5N and they did look good side by side with 21mpx ones from a 5D MKII

24mpx pictures with good per pixel resolution and increased DR are more than enough for my actual needs

[Lou Jost]

Is the single shot the first one of the 8 shoot series?

I don't know. But if any of the intermediate shots were blurred from motion, the algorithm would not work. Surely Chris_ma is right that they do not vary.

How long does it take each 8 shoot PS vs a normal shot?

Single shots take less than 1s, the minimum step time for most controllers. Shifted shots depend very strongly on the write speed of the card. On a fast Sony G series XQD card, the whole operation (collecting and combining the sub-shots, and writing to the card, and resetting for the next shot) takes 17-18s. On an SD card it takes 25s. Add to each of these times 1+ seconds to move the rail or turn a fine focus knob.

20s seems like a lot, maybe not something I would use for very deep stacks

Definitely impractical for deep stacks.

not worth the time and effort required to stack with Pixel-shift, especially if the final destination of the photo is the web

For web work even the cheap point-and-shoot or bridge cameras are adequate, and they can be used with microscope objectives. I sometimes use a cheap little Panasonic FZ300 fixed-lens bridge camera with a microscope objective when high resolution is not needed. I often also use MFT cameras for normal use.

For someone on a tight budget, the Panasonic bridge cameras (which shoot RAW) are wonderful solutions. No need to even buy a lens. But sometimes we have big plans for some of our images...and it is better to have un-needed resolution than to have not enough resolution, when that time comes.

so far I've seen a lot of comparisons and demonstrations about how great pixel-shift is, yet very few (if any) actual piixel-shift stacked photos at a resolution that allows to appreciate the benefits.

At the resolution of this forum, it is impossible to appreciate the benefits of these shifted images. But it does make a difference for medium to large prints. That's why medium format exists. This is a cheap way to get similar (or even better) quality. If you have ever been to an exhibit of large prints taken with Hasselblad or Phase 1 sensors, you have seen the quality difference. I want that too, but can't spend many tens of thousands of dollars. Now I get that quality for a relatively small investment.

[Macrero]

At the resolution of this forum, it is impossible to appreciate the benefits of these shifted images. But it does make a difference for medium to large prints. That's why medium format exists. This is a cheap way to get similar (or even better) quality. If you have ever been to an exhibit of large prints taken with Hasselblad or Phase 1 sensors, you have seen the quality difference. I want that too, but can't spend many tens of thousands of dollars. Now I get that quality for a relatively small investment.

At the forum resolution it's impossible to assess anything in a full frame picture, but you can always add an external link at higher resolution though.

I do not deny the benefits of pixel shift, they are there. But IMO for amateurs and enthusiasts, who takes photos as a hobby and share them on the web, pixel-shift is not really worth the time and effort. Unless you have plenty of free time and/or you can share the photo at a large enough size. Even in prints the difference is not that great, unless you print wall murals...

*I am referring to pixel-shift stacking. "Traditional", single shot photography is another story.

[Lou Jost]

It would be interesting to compare a pixel-shifted image with a stitch of four normal images. We'd lose the dynamic range improvement (gained from averaging multiple shots at each pixel) and but would probably improve the resolution, though probably with some moire. Using a standard controller and a hundred-image stack, four normal stacks (with very little overlap) would take 800s while a pixel-shifted stack would take 2600s. If we used focus bracketing of the lens to generate the stacks, the four normal stacks could be done in much less than 400s.

Unless you have plenty of free time and/or you can share the photo at a large enough size. Even in prints the difference is not that great, unless you print wall murals...

People have tested viewers' preferences for medium format vs 35mm prints at various sizes. I don't remember the details, but the medium format advantage was clear even in relatively moderate print sizes. That's why medium format exists. It is not just for wall-sized murals. But you are right that if this is just something people are doing for fun, with no desire to print large, it is a waste.

[Macrero]

Print quality will depend on various factors. The number of pixels matters, of course. I don't believe there is such thing as "medium format look" though. Not to mention how difficult (and expensive) it would be to work in a wide magnification range with a MF camera.

If you need high resolution images, then pixel-shift is an excellent option and a good and "cheap" alternative to medium format cameras. But it is hard to justify its use in stacking if the purpose of the photographs is web-sharing (even at a "reasonable" size) or relatively small/medium-sized prints. That's why in order to clearly display the difference you need to show 200% or 400% crops. The difference in prints would be even smaller.

That's nothing but my personal opinion, of course.

[Beatsy]

Print quality will depend on various factors. The number of pixels matters, of course. I don't believe there is such thing as "medium format look" though. Not to mention how difficult (and expensive) it would be to work in a wide magnification range with a MF camera.

If you need high resolution images, then pixel-shift is an excellent option and a good and "cheap" alternative to medium format cameras. But it is hard to justify its use in stacking if the purpose of the photographs is web-sharing (even at a "reasonable" size) or relatively small/medium-sized prints. That's why in order to clearly display the difference you need to show 200% or 400% crops. The difference in prints would be even smaller.

That's nothing but my personal opinion, of course.

A fair summary.

16-shot shift seems occasionally useful to squeeze a little extra resolution (detail) that doesn't *quite* make it through due to image scale - and then gets mangled by the Bayer decode. The results are even useful for web publishing (with 100% cropped insets) but other than that, I'll mostly use it for single images in future, not stacks. The 4-shot stacking (for true R,G,G,B colour at each pixel) is a different story though. It's much more tolerable to use (4x more tolerable, in a way ) and the subsequent benefit is proportionally greater than for 16-shot IMO. It works at pixel level but the effect brings the whole tone and dynamic range of an image closer to that "medium format look" IMO - even without boosting the scale to three-figure megapixels. This holds true in all photography contexts (micro, macro and meso) so I'll definitely be using 4-shot pixel shift a lot from here on in. On the microscope in particular - it's brilliant there.

[Macrero]

I've owned 3 cameras with pixel-shift, none of them had the 4X "Sony" approach, so can't comment on that.

Checking out the external hard drives, I found this comparison I made with the S1R. Nikon S Fluor 4/0.20 + Sigma LSA, 100% crop:

http://images2.imgbox.com/d6/c2/T9n9DUCv_o.gif

The difference is clear, isn't it? But trust me, depending on the subject you would be very hard-pressed to tell the difference in a print.

Found this one as well. Oly 10/0.30 + Ysaron 153.

HR, 100%:

https://images2.imgbox.com/ea/f2/cspl2cBx_o.jpg

Single shot upsized, 100%:

https://images2.imgbox.com/4d/7a/AXH0zgfY_o.jpg

Same as above, and even less if you sharpen the single shot image.

And the S1R was the the camera I saw biggest difference with. The difference with the MFT cameras was even smaller.

Moral of the story? The perfect camera or tech still does not exist. If you are an experienced photographer looking for the last bit of resolution/quality, try it and decide for yourself if it's worth it or not. If you are are starting up with photomacrography, don't rush to get a gearhead and resolution freak and aim to improve your technique, lighting, composition... that will make a much bigger difference than shifting pixels.

[Lou Jost]

I don't believe there is such thing as "medium format look" though. Not to mention how difficult (and expensive) it would be to work in a wide magnification range with a MF camera.

I agree that some aspects of the alleged "medium format look" don't exist. But other aspects (DR and smoothness of color gradations) are as real as the corresponding difference between a full frame photo and an MFT photo. Of course you are right that medium format would be an expensive solution for macro work, but that's exactly why it is so exciting to be able to get that degree of DR and subtlety of color gradations, without any additional investment, from our existing cameras.

I may be biased as I used to do 4x5 large-format landscapes and macro, and those photos just blew away 35mm film. I love immersing myself in a photo's details. I definitely don't buy the common claim that there is a standard print-observing distance which makes fine detail unnecessary in large prints.

[Lou Jost]

I've owned 3 cameras with pixel-shift, none of them had the 4X "Sony" approach, so can't comment on that.

It's worth mentioning that the Pentax K-1 also has this kind of four-image pixel shift. It makes a big difference over single shots.