Anyone trying pixel-shifting camera scanning?

[GLS]

More to chew on! Another set of copies of a 35mm Tri-X negative, this time using an A7RIV with Sigma Art 70mm Macro, set at f/4. From top to bottom, these are single-shot, four-shot and sixteen-shot captures. The multi-shot captures were converted to single DNG files using the LibRaw PixelShift2DNG utility. Raw files were converted in RawTherapee 5.8 and saved as TIFF files, then opened in Picture Window Pro 8 for final processing, which consisted solely of upressing the smaller files using bicubic with sharpening set to 0, and then cropping for presentation. Again, the crops are presented here at 100%.

This clearly shows the cleaner grain representation of the 16-shot version compared with the "worms" from the single shot.

Really though, the benefits for 35mm Tri-X are going to be extremely marginal compared with larger formats and better resolving films.

 

[Oren Grad]

Really though, the benefits for 35mm Tri-X are going to be extremely marginal compared with larger formats and better resolving films.

I've gone down this rabbit hole because faithful rendering of Tri-X grain matters to me - for my purposes it's an important part of image character independent of the degree to which the subject in the picture is resolved. "Quality" is a tricky word in this context. If I want the utmost in technical quality I use large format and make contact prints in the darkroom. However, more generally I think not so much of "quality" as "qualities" - the different attributes of a picture that give it its distinctive flavor. From that perspective 35mm Tri-X is a medium in its own right, and the grain is part of its character.

But there's no one right answer here. We each have our own esthetic tastes and photographic goals, and how demanding one needs to be about scanning depends on subjective judgments about which aspects of image character matter for our respective purposes.

As a practical matter, there is a steep price in equipment cost and workflow hassle for this extra fidelity. The effort and file overhead involved are such that, at least for my purposes, using high-resolution cameras in this way is completely impractical for routine scanning of entire rolls. Fortunately, I don't need that - as in the darkroom, I select particular frames in which I think I want to invest effort for possible printing, and copy only those.

EDIT: I should add, even when copying selectively, I don't pull out all the stops all the time. When I'm copying 35mm, a single GFX 100S shot is enough for my purposes most of the time. I haven't done much copying of medium format, but if I were, I'd be using multi-shot more often. The alternative is stitching, which IMO is even more hassle. But of course preferences about this vary too.

 

[McDiesel]

@4season Resolution is the most boring and least relevant aspect of scanning because the plateau of diminishing returns makes it so. Even a budget camera with an OK lens will let you cover the distance to that plateau instantly. For every subsequent improvement after that you'll be paying with time, sweat, dollars, etc. In other words, camera scanning lets you reap 90% of what's possible with zero effort.

You can cover additional 5% with a premium solution such as Sony a7R IV, but additional resolution gain will be barely noticeable and bloody. Even the pixel shifting tech is not without caveats.

Oren's samples above illustrate how pointless this is - any microscopic additional detail is drowning in grain anyway.

The additional resolution gains go towards improving grain quality. It makes perfect sense to me because it's the combination of a film camera taking lens and the grain that set the practical limit for enlarging. That is why I am not sure that extracting the remaining 5% of data from film is worth it. Apparently the world doesn't think so, because I have never seen a scan (from any device) that beats a 60MP pixel-shifting Bayer sensor with a top of the line flat field lens. This web site is a good source for those who want to try.

 

[faberryman]

For what purpose are you making camera scans of your negatives?

 

[4season]

@4season Resolution is the most boring and least relevant aspect of scanning because the plateau of diminishing returns makes it so. Even a budget camera with an OK lens will let you cover the distance to that plateau instantly. For every microscopic improvement after that you'll be paying with time, sweat, dollars, etc. In other words, camera scanning lets you reap 90% of what's possible with zero effort.

You can cover additional 5% with a premium solution such as Sony a7R IV, but every additional microscopic improvement will be bloody. Even the pixel shifting tech is not without caveats.

Oren's samples above illustrate how pointless this is - any microscopic additional detail is drowning in grain anyway.

The additional resolution gains go towards improving grain quality. It makes perfect sense to me because it's the combination of a film camera taking lens and the grain that set the practical limit for enlarging. That is why I am not sure that extracting the remaining 5% of data from film is worth it. Apparently the world doesn't think so, because I have never seen a scan (from any device) that beats a 60MP pixel-shifting Bayer sensor with a top of the line flat field lens. This web site is a good source for those who want to try.

Very interesting site, thanks for the link.

This is very much about me geeking out about what's possible (in terms of lines/mm), without much thought about what I'd actually do with it. And I've already paid for toys long ago: 2 pixel-shifting camera bodies, 4 macro lenses, some Tmax 100 film in the fridge, even a lens salvaged from a Microfiche reader.

As for highly-magnified grain, it seems to have it's uses:
https://www.photrio.com/forum/threads/hiw-did-masahisa-fukase-achieve-this-effect.184992/

 

[bags27]

Very interesting site, thanks for the link.

This is very much about me geeking out about what's possible (in terms of lines/mm), without much thought about what I'd actually do with it. And I've already paid for toys long ago: 2 pixel-shifting camera bodies, 4 macro lenses, some Tmax 100 film in the fridge, even a lens salvaged from a Microfiche reader.

As for highly-magnified grain, it seems to have it's uses:
https://www.photrio.com/forum/threads/hiw-did-masahisa-fukase-achieve-this-effect.184992/

Feel the same. I'm already in for a pound; I might as well be in for the penny.

And while I'm not sure I want that much grain as the link , a big reason for my choosing Tri-X is that, despite it changing its old formula long ago, it still references old school film and I want to pick up that grain.

 

[McDiesel]

@4season Hehe I hear you. Actually, I am contemplating ordering a microscope with a camera port, to see what's possible. It is probably the cheapest and most convenient way to reach the absolute limits. The reason I'm hesitating is that I know it will be gathering dust and taking up space after a weekend of experiments.

 

[Lachlan Young]

there's no extra resolution to be revealed... However, the grain quality goes up substantially when stiching or using pixel shift.

The quality of the resolution of visual granularity is what matters when discussing scan quality, rather than questionable 'debate' over image information held on the film. Too many people get totally distracted by making category error claims about limits of of film resolution, without realising that it's the ability to reproduce both the information held on the film and the granularity quality of the film that together create our perception of good resolution at larger and larger sizes. In other words, a convincing 25x off HP5+ requires not just that you properly and sharply transmit the information contained by the film, but enough of the film's own inherent visual character too. This is where specialist enlarging lenses come into play - and where assumptions about scanning resolutions can get questionable (assuming the scanning system is adequately sharp in the first place).

 

[Adrian Bacon]

I currently use my Fuji GFX-R (MF 50 mps) with a Mamiya 625 f/4 120mm macro to scan my MF (6x6, 6x7, and 6x9) negatives. Works well. But always looking for better.

I read about a Large Format shooter who so liked his pixel-shifting Olympus micro 4/3 for scanning (I imagine he stitched several) that he actually got rid of his drum scanner. Wow! That's quite a claim.

I could swap my Fuji for a used pixel-shifting Panasonic S1R (FF, 47 mps) to pair with my Sigma L mount macro lens. I'm thinking of renting it first to compare. But just wondering if anyone's tried this technology for film scanning.

Of course, if that's great, the new pixel-shifting Fuji GFX 100-S might be even better, but that's not being discounted from $6k, and there has to be a limit to how effective larger sensors are going to be, right?

thanks!

Meh... There's definitely a point of diminishing returns at least to human eyes. Sure on paper it sounds superior, and if you zoom in and pixel peep at 800% I'm sure you could see differences, but in reality, 20-30MP bayer sensor is more than enough resolution for literally 90%+ of the output cases. You can easily make high quality prints up to 16x24 with 24MP, and I can personally attest that the film grain of pretty much every emulsion you can get in 35mm is very visible at 6000x4000 pixels, and it isn't because of grain aliasing because the cameras I use for scanning have AA filters in front of the sensor.

So, can you do it? Sure. Is it worth the extra effort over a single capture? Probably not, unless you had a really specific use case or regularly printed extremely large, in which case, you would know how much resolution you actually needed and would shoot an appropriate format for that output.

 

[JerseyDoug]

The quality of the resolution of visual granularity is what matters when discussing scan quality, rather than questionable 'debate' over image information held on the film. Too many people get totally distracted by making category error claims about limits of of film resolution, without realising that it's the ability to reproduce both the information held on the film and the granularity quality of the film that together create our perception of good resolution at larger and larger sizes. In other words, a convincing 25x off HP5+ requires not just that you properly and sharply transmit the information contained by the film, but enough of the film's own inherent visual character too. This is where specialist enlarging lenses come into play - and where assumptions about scanning resolutions can get questionable (assuming the scanning system is adequately sharp in the first place).

Exactly. That's why we used grain magnifiers, not image magnifiers, back in the darkroom days. (And of course many still do today.)

 

[Empyreus]

Here are three captures from the same 35mm Tri-X negative. The top one is Nikon LS-9000, the middle one is GFX 100S with 75mm f/4 Apo-Rodagon-D 1:1 lens, and the bottom one is GFX 100S in pixel-shift mode with 75mm f/4 Apo-Rodagon-D 1:1 lens. For both of the GFX captures the Apo-Rodagon-D was set to f/5.6 marked aperture.

The GFX raw files were processed in Capture One with sharpening set to 0 and saved as TIFF files. All three TIFF files were opened in Picture Window Pro 8 for final processing. Minor curve adjustments were made to roughly match contrast and brightness to minimize discrepancies that might bias perception. The two smaller files were upressed to match the pixel-shift file using bicubic with sharpening set to 0. The crops here are presented at 100%.

Do you feel that the GFX100S is a worthwhile upgrade from a standard fuji mirrorless in regards to scanning? Always looking to improve my scans and the medium format Fuji have been interesting me.