Scanning with a camera supporting pixel shift

McDiesel · Jul 3, 2022

I can't find the original thread for some reason, but someone was asking for full-sized scan samples from a camera supporting pixel shift. I finally got around to this, although I only have HP5+ unfortunately, so in my recent roll I found an (extremely rare!) image that actually contains usable detail to serve as an example.

Here we go, full-sized 12,000 x 12,000 pixels scan of an 6x6 negative.

Details:

Sony A7R IVa, 16-shot mode
Sigma 105mm f/2.8 DG DN Macro Art Lens set to f/5.6
Lightroom sharpening is set to radius of 1, amount 55, detail 25
HP5+ exposed with Hasselblad 100mm Planar CFi on a tripod at f/8
Developed in Ilfotec HC, 1+47 dilution

Personal observations

Dealing with the files of this size (50MB+) is hugely impractical. Notice how much it takes for your browser to render it? Here's my "working scan", i.e. the kind of scan that I would normally keep. It's 75% smaller in size and demonstrates the law of diminishing returns quite nicely.
The lens can't keep up towards the edges, even for the square (!) And this is the sharpest native-mount AF lens I found for this mount.
I had to wait for everyone in the house to fall asleep to avoid micro-vibrations. 16-shot mode won't work on a 2nd floor in a "sticks+cardboard" typical American house if someone else is moving.

[EDIT] to avoid your browser freezing, right-click on the link and select "Save as".

gone · Jul 3, 2022

Locked up my laptop twice clicking on that file. Had to remove the battery and do a hard reboot to clear it out too. Never had that happen here. Good luck w/ it, I won't click on it again.

bags27 · Jul 3, 2022

thanks so much! Lovely photo!!!

It may have been me who made the request: I've certainly been raising the question here and elsewhere.

I currently scan with a Fuji GFX-R and a Mamiya 645 120 f/4 macro. I'm getting files of around 100mb that often end up in post to be 3 or 4 times that. So I'm used to big files. (My computer handled this download easily). And I scan at night using a massive Beseler enlarger as my copy stand. So, I think I've taken scanning about as far as I can, until...

I do think your scan may be a tad sharper than what I produce. (By the way, just wondering if you determined whether the Sigma is sharpest at f/5.6; my 70mm copy is sharpest at f/8.)

Anyway, thanks so incredibly much. Thinking I may have to swap out my GFX-r for a pixel-shifter or at least rent one first and do a comparison.

McDiesel · Jul 3, 2022

bags27 said:
I do think your scan may be a tad sharper than what I produce. (By the way, just wondering if you determined whether the Sigma is sharpest at f/5.6; my 70mm copy is sharpest at f/8.)

I did not notice any difference myself between f/5.6 and f/8 but there are in-depth reviews of this lens online that suggest a slight effect of diffraction at f/8, and I could use the extra stop of speed when digitizing. I found vibrations and negative flatness to be the bigger enemies than optics.

There's another rabbit hole to explore: focus stacking, when you can do 4 exposures focusing on the center + 4 edges or corners and then stack them. Another one is stitching, but negative flatness makes that less effective. TBH I just don't have the energy for this jazz. I only scanned this shot this way out of curiocity. My regular method of digitizing is 4-shot pixel shift without stitching or stacking for medium format downsampled to 6000px wide (sample), and single-shot for 35mm. About 5 minutes per roll.

BTW I tried digitizing with the GFX 100s, and found the lack of AF to be the bigger bottleneck than anything else. It's really hard to be consistent with manuall focus vs Sigma AF.

McDiesel · Jul 3, 2022

And just for fun, here's 100% crop from the 250MP+ scan using pixel shift and stitching at the same time.

bags27 · Jul 3, 2022

McDiesel said:
I did not notice any difference myself between f/5.6 and f/8 but there are in-depth reviews of this lens online that suggest a slight effect of diffraction at f/8, and I could use the extra stop of speed when digitizing. I found vibrations and negative flatness to be the bigger enemies than optics.

There's another rabbit hole to explore: focus stacking, when you can do 4 exposures focusing on the center + 4 edges or corners and then stack them. Another one is stitching, but negative flatness makes that less effective. TBH I just don't have the energy for this jazz. I only scanned this shot this way out of curiocity. My regular method of digitizing is 4-shot pixel shift without stitching or stacking for medium format downsampled to 6000px wide (sample), and single-shot for 35mm. About 5 minutes per roll.

BTW I tried digitizing with the GFX 100s, and found the lack of AF to be the bigger bottleneck than anything else. It's really hard to be consistent with manuall focus vs Sigma AF.

Thanks. I agree. I've found that stitching is just too dreary unless one has an unbelievable shot and is willing to keep working at it. Does focus stacking improve an image of a flat object? I didn't realize that!

Thanks again!

bags27 · Jul 3, 2022

McDiesel said:
And just for fun, here's 100% crop from the 250MP+ scan using pixel shift and stitching at the same time.

wow. That IS fun. Pretty impressive, actually.

McDiesel · Jul 3, 2022

bags27 said:
Does focus stacking improve an image of a flat object? I didn't realize that!

It does not, quite the opposite. Focus stacking allows you to easily deal with never-quite-completely-flat negatives. There are even devices that automate this process.

bags27 · Jul 3, 2022

McDiesel said:
It does not, quite the opposite. Focus stacking allows you to easily deal with never-quite-completely-flat negatives. There are even devices that automate this process.

Such a great point: thanks!

Adrian Bacon · Jul 3, 2022

McDiesel said:
I can't find the original thread for some reason, but someone was asking for full-sized scan samples from a camera supporting pixel shift. I finally got around to this, although I only have HP5+ unfortunately, so in my recent roll I found an (extremely rare!) image that actually contains usable detail to serve as an example.

Here we go, full-sized 12,000 x 12,000 pixels scan of an 6x6 negative.

Details:

Sony A7R IVa, 16-shot mode

Sigma 105mm f/2.8 DG DN Macro Art Lens set to f/5.6

Lightroom sharpening is set to radius of 1, amount 55, detail 25

HP5+ exposed with Hasselblad 100mm Planar CFi on a tripod at f/8

Developed in Ilfotec HC, 1+47 dilution

Personal observations

Dealing with the files of this size (50MB+) is hugely impractical. Notice how much it takes for your browser to render it? Here's my "working scan", i.e. the kind of scan that I would normally keep. It's 75% smaller in size and demonstrates the law of diminishing returns quite nicely.

The lens can't keep up towards the edges, even for the square (!) And this is the sharpest native-mount AF lens I found for this mount.

I had to wait for everyone in the house to fall asleep to avoid micro-vibrations. 16-shot mode won't work on a 2nd floor in a "sticks+cardboard" typical American house if someone else is moving.

[EDIT] to avoid your browser freezing, right-click on the link and select "Save as".

This is interesting, though I’d like to see a normal scan of the exact same frame as a comparison.

McDiesel · Jul 3, 2022

@Adrian Bacon here. slightly downsampled though, it was a bit over 6000px.

Alan Edward Klein · Jul 3, 2022

bags27 said:
Thanks. I agree. I've found that stitching is just too dreary unless one has an unbelievable shot and is willing to keep working at it. Does focus stacking improve an image of a flat object? I didn't realize that!

Thanks again!

How do you focus stack a negative?

McDiesel · Jul 3, 2022

@Alan Edward Klein just like any 3-dimensional object: you take multiple photos focusing on different parts of it, and then you merge them in Photoshop. There are blogs and video tutorials online.

This is better than trying to sandwich negatives with glass, because in addition to dealing with negative (non)flatness it also helps correct for field curvature of your lens.

Adrian Bacon · Jul 3, 2022

McDiesel said:
@Adrian Bacon here. slightly downsampled though, it was a bit over 6000px.

Thank you.

I have to say, the pixel shift image clearly renders the grain in the sky with more detail than the single capture image, but zooming into the top of the mountain and looking at the trees and buildings that are there, the single capture seems to be sharper and have more detail than the pixel shift image, which seems weird to me as looking at the sky, the pixel shift image clearly has sharper and more defined grain.

I pulled both into Photoshop, enlarged to 400% on the single capture, and 200% on the pixel shift.

I may also crop both to 2:3 aspect, then scale them down to a 12x18 print at 300 pixels per inch and look at them that way too.

I've always been on the fence about pixel shifting and whether it delivers real, usable detail when scanning film, and this is purely me, but I have to say, this so far is kind of a wash. I can see where the grain is sharper with more detail there, but at the same time, there are other aspects that I'd have expected to have more detail and are actually softer. Maybe this is illustrative of the various technical challenges that comes with pixel shifting, which has me even more on the fence, as it's potentially even more fiddly than is worth it except for special images where you're going to put the effort in to get there.

McDiesel · Jul 3, 2022

@Adrian Bacon I just realized that the smaller image I shared also used pixel shift, but in 4-shot mode (same image size but no bayer interpolation), sorry. I don't have a regular single-shot of that negative to share, maybe I'll find the time later to make one.

Adrian Bacon · Jul 3, 2022

McDiesel said:
@Adrian Bacon I just realized that the smaller image I shared also used pixel shift, but in 4-shot mode (same image size but no bayer interpolation), sorry. I don't have a regular single-shot of that negative to share, maybe I'll find the time later to make one.

OK. I thought my eyes were playing tricks on me, either that or the single shot one had sharpening applied. Either way, I'm very interested in seeing single shot, relative to 4 shot and 16 shot. I'd say the 4 shot one overall is the sharper one when cropped and scaled to 12x18 print size at 300 pixel per inch.

McDiesel · Jul 3, 2022

I'll see if I can find time later today. I'll post DNGs so you'll be able to play with your own sharpening.

McDiesel · Jul 4, 2022

@Adrian Bacon here we go:

16-shot pixelshift DNG
Normal "bayer" DNG (one of the 16 images that the previous DNG was composed of)

This way they're 100% identical (focus, alignment, etc). This allows you to play with your own sharpening preferences, which as you already know has a huge impact on the grain. To my eye there's clear resolution boost by looking at the trees on the top, even though the pixel-shifted image shows a bit of micro-vibration side effects. But the grain quality is notably higher. This is HP5+ after all.

And here's the final product with some sharpening applied. This is what I would be keeping in my albums (minus dust), with the DNG going to AWS Glacier.

Adrian Bacon · Jul 4, 2022

McDiesel said:
@Adrian Bacon here we go:

16-shot pixelshift DNG

Normal "bayer" DNG (one of the 16 images that the previous DNG was composed of)

This way they're 100% identical (focus, alignment, etc). This allows you to play with your own sharpening preferences, which as you already know has a huge impact on the grain. To my eye there's clear resolution boost by looking at the trees on the top, even though the pixel-shifted image shows a bit of micro-vibration side effects. But the grain quality is notably higher. This is HP5+ after all.

And here's the final product with some sharpening applied. This is what I would be keeping in my albums (minus dust), with the DNG going to AWS Glacier.

28MB is awfully small for a 60+ MP DNG for the normie file. Exiftool gives warnings trying to read it and all of the IFD and SubIFD chunks are 8 bits and jpeg compression. That doesn't seem right... When doing multishot, it doesn't save raw files?

Helge · Jul 4, 2022

As you say HP5 is really not ideal for this. It’s far more grainy and lower resolving than D100 or TMX.
Super for pushing and dynamic range, but for this?

McDiesel · Jul 4, 2022

@Adrian Bacon they are compressed DNGs converted from Sony ARW/ARQ by Lightroom. It's just impractical for me to keep half-a-gig raw files sorry.

@Helge actually, HP5+ is a better patient for high-res scanning because of grain. I find grain to be a more visible indicator of a poor scanning, even at a small magnification/reproduction. This photo captured all the detail I needed and then some, that's why I almost never use slower films in medium format.

bags27 · Jul 4, 2022

I found that with a little work in Camera Raw, I could make the "normal" shot look quite competitive. And getting rid of the sky grain/noise isn't really much of a problem if I wanted to do it.

Of course this is only n=1....and my knowledge/brain power is way below Adrian Bacon's to appreciate that significance. So, unless there's something vastly skewed about this comparison, I think this has scratched my itch to try pixel-shifting scanning. Doesn't seem to be a magic bullet, or even a significant incremental difference.

Thanks so much!

Adrian Bacon · Jul 4, 2022

McDiesel said:
@Adrian Bacon they are compressed DNGs converted from Sony ARW/ARQ by Lightroom. It's just impractical for me to keep half-a-gig raw files sorry.

@Helge actually, HP5+ is a better patient for high-res scanning because of grain. I find grain to be a more visible indicator of a poor scanning, even at a small magnification/reproduction. This photo captured all the detail I needed and then some, that's why I almost never use slower films in medium format.

I don’t care about the compression or no compression, what I care about is the bit depth. none of the dng IFD chunks have the camera original 14 bit samples. Not having that fidelity can hurt and introduce posterization and banding when you start torking the samples around to get to a linearized positive image. The film grain will act as a dither and mask it to some degree, but at some point depending on how hard you step on it, it’ll show up because you only have 255 discrete tone values that you’re stretching out to at least 12-13 stops of scene DR.

McDiesel · Jul 4, 2022

Heh... well, I already deleted the uncompressed RAWs. Do you see what you wanted to see in the DNGs?

Lachlan Young · Jul 4, 2022

If you want a further comparator, I can throw a 6300ppi scan of 120 HP5+ (or one of the 3 strips at least from before I stitched them) into the mix from the Hasselblad/ Imacon implementation of the Kodak 3x CCD linear sensor - and handled so as to minimise Flexcolor's sometimes interesting idea of inversion etc.

Scanning with a camera supporting pixel shift

Privacy & Transparency

Privacy & Transparency