DSLR imaging problem

[Adrian Bacon]

After some testing, I can confidently say that film curl sucks

Basically, the pixl-latr, which does fine on medium format and 4x5, doesn't constrain 35mm as much as I would like. As soon as I put the film into the Epson holder, and used that for imaging, the focus came right back in.



I've had this lens for a very long time-- it was the first "quality" lens I bought for myself (ie, one where the review didn't say "It's a good lens for the price..."). Didn't realize the resolution was quite that weak on APS-C, and of course, the 90D is going to push that even harder.

It’s not a terrible lens, it’s quite good in fact, it’s just better suited to full frame cameras because it maintains almost the same performance that it has in the center all the way out to the edges, which for full frame is awesome. That lens on a 5DSr is awesome. There are just newer lenses that render higher resolution in the center of the frame, which is better suited for an APS-C sensor. Dxomark is your friend. You can select a lens then select an APS-C camera that they tested it on and look at the effective MP it rendered on that camera. Then select the 5DSR for the same lens and look at the difference. There are definitely lenses that render a lot more res in the center of the frame.

all that being said, yes, being in focus is more important. Unless I had a way to keep the film completely flat, a much higher DOF is more desirable. Again, it’s a balancing act between DOF, diffraction, and vibration, as the more you stop down, the longer your shutter time unless you have a lot of light.

 

[grat]

It’s not a terrible lens, it’s quite good in fact, it’s just better suited to full frame cameras because it maintains almost the same performance that it has in the center all the way out to the edges, which for full frame is awesome. That lens on a 5DSr is awesome. There are just newer lenses that render higher resolution in the center of the frame, which is better suited for an APS-C sensor.

Oh, understood-- in fact, the lens still produces some very nice results for general photography. As I mentioned above, the lens is fine out to the region of 50 lp/mm, at which point it starts dropping off, especially away from the center. That's really only going to be an issue when pushing limits like I am with trying to image film at > 4800 PPI.

Dxomark is your friend. You can select a lens then select an APS-C camera that they tested it on and look at the effective MP it rendered on that camera. Then select the 5DSR for the same lens and look at the difference. There are definitely lenses that render a lot more res in the center of the frame.

Ha!! They have so far refused to do any testing of the 32.5MP APS-C Canon sensor on either the M6 Mk II or the 90D. I know they've tested other sensors, but as far as I know, the 90D still has a remarkably dense sensor.

all that being said, yes, being in focus is more important. Unless I had a way to keep the film completely flat, a much higher DOF is more desirable. Again, it’s a balancing act between DOF, diffraction, and vibration, as the more you stop down, the longer your shutter time unless you have a lot of light.

Light isn't so much a problem. I was shooting at f/8 and 1/30th (ISO 100). I could probably raise the ISO to 200, and shoot at f/11, as long as I ran the RAW file through DPP4.

 

[Adrian Bacon]

Ha!! They have so far refused to do any testing of the 32.5MP APS-C Canon sensor on either the M6 Mk II or the 90D. I know they've tested other sensors, but as far as I know, the 90D still has a remarkably dense sensor.

This is true, but, it's not dramatically denser than the 80D or any other 24MP APS-C sensor. The difference is 6000 pixels along the long edge vs 6900 pixels along the long edge. Not a massively huge difference. 24MP to 32MP sounds like a huge step up, but in practice, it's not nearly as big of a difference as it sounds like.

 

[Adrian Bacon]

@Adrian Bacon but don't you think this advantage goes away with dedicated reduced-circle lenses made specifically for APS-C, like Fuji's X or Canon's EF-S? Perhaps a better approach is to scan with a dense FF sensor but zoom out a bit to keep the image away from the corners? I am contemplating building a rig based on Fuji's 100MP sensor to avoid stitching MF scans, and wonder how this approach would work for 35mm scans...

I'm not sure I understand what you mean. Typically, the bigger the imaging area and the bigger the lens image circle is, the less performant it is over the whole area. With film, this is OK, because each film size is a really big jump up from the next one, so even though you lose 30% performance (as an example), you more than quadrupled your film area size, so the total resolution is still dramatically higher with the next size up in film. With digital sensors, quadrupling the imaging area is not feasible or it would have already been done. So camera makers have gone the other way and kept basically the same imaging area (35mm and APS-C) and just really pushed sensel density and lens design.

The 90D is the densest pixel density of any sensor made by pretty much anyone with the same or better performance than Canon has been getting. If you were to take that pixel density and blow it out to full frame size, you'd be looking at ~11200x7466 pixels (or ~84MP for full 24x36mm sensor size). I doubt Canon's own new RF glass could keep up with that across the full 24x36 frame, but... Canon and Sigma both make glass that is pretty sweet and does a really good job of keeping up with an APS-C sensor with that density in the center of the imaging area. There are a couple APS-C lenses that do as well, though they're not macro lenses.

You could certainly go with a much larger medium format sensor that has a much lower pixel density, and it'd work really great for scanning 120 and 4x5 or larger film, but at 1:1, scanning a full 24x36mm piece of film, you'd be getting quite a bit less resolution for the 35mm film frame. Yes, you'd be out of the corners of the lens, but you'd be pretty hard pressed to get more resolution than the 90D and a full frame lens with good center resolution.

APS-C is kind of the sweet spot for scanning film. It's high enough resolution with lenses that are easy enough to make performant without spending a fortune, and full frame cameras aren't quite as big of a jump up in resolution even though the sensor is a lot larger, the lenses don't typically render as high out to the edges. The focus field tends to be a lot flatter in the center of the frame with APS-C sensors too, which is another plus. Typically, the bigger the imaging area and the bigger the lens image circle is, the less performant it is over the whole area.

 

[Adrian Bacon]

Ah, makes sense now. Thanks.

Yep.

As another example, it's relatively easy to jam 100+ line pairs per mm in an image circle that covers APS-C, and for full frame, if you spend a lot of money, you can get there, but usually you're less than 100lp/mm. Once you get up to 6x6 sized image circle, an average lens is down in the 60-70lp/mm range with really good and expensive glass maybe hitting 100lp/mm. 4x5? You're doing good if you can manage 60lp/mm+ over the whole frame, same for 8x10.

Now in terms of total resolution, APS-C is easily 20+ MP with a good lens. Full frame can improve on that, but not by a lot. 6x6? Even with an average run of the mill lens, you're *easily* in the 40+MP range. 4x5? Even with a crappy lens that does a measly 40lp/mm you're popping out 80+MP, with 8x10 at least 4 times that resolution.

So, yeah. I probably wouldn't go smaller than APS-C for scanning, and full frame can be useful for 120 and larger, but outside of that, spend the money on a macro lens that performs in the center and get a high resolution APS-C body to go with it.

 

[Adrian Bacon]

Just looking up some specs.... I didn't even realize beasts like 90D exist. I just assumed that all current-gen sensors have the same pixel density, i.e. essentially different wafer cuts from the same Sony fab. Fuji's 100MP GFX 100S is just more of the same 3.7 micron pixels as Fuji X-T3. Totally forgot that Canon makes their own sensors.

Canon gets a lot of heat because their sensors don't "perform" with high DR relative to their competition, but in reality, they prefer to keep the same ~12-13 stops of DR and just push the pixel density up. Sony and Nikon both are dynamic range kings, but the tradeoff has been less resolution. Canon was first (by huge time margins) to 18MP, 20MP, 24MP, and 32MP in APS-C land, and was first to 20+MP, and 50MP in full frame land.

Sony is just now barely exceeding the resolution with their A7R4 that Canon has been at with the 5Dsr for how many years? At least 5. Yeah, it's been that long. Everybody has been oohing and ahhing over dynamic range with Sony and Nikon because they seem to think they can't take good photos with less than 14-15 stops, and in the meantime, Canon has just been keeping the same totally usable ~12ish stops and quietly dropping complete beast mode resolution monsters on the market. The first thing reviewers inevitably say is "the dynamic range sucks", and make minor mention of class leading resolution, and completely omit that if you can't take any good photos with 12 stops of DR, then it's not the camera. Meanwhile, working professionals that aren't prone to switching camera systems every other day, simply upgrade when funds permit and welcome the resolution boost.

Rumors floating around is their next step up will be 80+ in full frame land, and who knows what it'll be for APS-C. The 32MP of the 90D already out resolves a lot of lenses.

 

[Adrian Bacon]

I am currently scanning 35mm with Canon 5D Mk4. FF sensor with 30MP and 100mm f/2.8 L macro lens. I lift the camera higher when scanning 35mm to leave some room around the edges, effectively getting only 6,000x4,000 pixels instead of the full 6,700x4,300, and then I downsample to 5,000px on the wide edge anyway, with slight sharpening. The results are spectacular and I feel I am not leaving any resolution on the table, even for ISO-100 films. But this setup is not ideal for 6x6 scanning. I stitch two frames, and the resulting 6,000x6,000 image doesn't seem to capture all available detail (easily proven if I lower the copy stand and take a closer shot), so I'm looking for the most cost-effective upgrade. The 100MP Fuji seems sweet but $$$.

120 is where flatbed scanners start to come into their own. Even with my 90D, I can't quite hit the 2400DPI that I can easily get with with my Epson V850Pro. I still scan with the 90D as it's way faster, but if I need that boost in resolution, then it goes on the flatbed. And with 4x5 or 8x10? DSLR is nice for a quick and easy proof, but even at pedestrian DPIs, the flatbed blows it out of the water.

I know the 5Dsr is getting really long in the tooth, but it's a pretty significant jump in resolution for 120 and you can use the same lens.

 

[Adrian Bacon]

V850 with stock film holders? And you probably used the Simple Tools to invert?

I've been debating a flatbed, but I have been turned off by the samples the V850 owners proudly display online. The colors are awful and often there are weird edge halo effects visible even in downsampled (!) images. Some look spectacular though, like our very own @Alan Edward Klein but that's an exception to the rule. So I can't figure out if it's a copy to copy variation or the mastery of Silverfast, or something else. I've had two scanners: the V600 and the Plustek 120 Pro and both couldn't approach the quality of my DSLR scans (same operator, obviously) - strong color casts that require massive post-processing efforts, shadow noise, dust, etc...

v850 against the platen glass with a sheet of anti-Newton glass on top, scan in 8x10 mode which focuses on the platen glass, vuescan to get a raw tiff file, and yes simple image tools. 35mm is just OK, but 120 and larger is quite good. I scan at the sensor native 4800dpi, and scale it down to 2400dpi. For black and white I scan monochrome using just the blue channel and dump it out in raw. It’s not a cheap scanner, and can be hamstrung by film holders and operator error, but, for my uses where I need more than 30MP or want to make an archival copy of something, it’s not a slouch.

 

[grat]

One issue I've mostly ignored in this thread is the color profile... for B&W, obviously, it's a non-issue. But one of the complaints Nick Carver had about DSLR scanning, which I've noticed as well, is the ... for lack of a better term, "personality" of film scans from the Epson vs. the Canon. His complaint was that film, whether it was Fuji 400H or Kodak Portra, came out looking like it was shot with a Canon DSLR (which, well, it was).

I can get a DSLR digitization to look somewhat like the colors of a scanned color negative-- but there's always a subtle bit of variation that shifts it from looking like film, to looking like "Canon". I suspect the Epson of having a more "neutral" bayer array and/or demosaicing algorithm, but these things are hard to quantify.

 

[grat]

But different films *do* have different looks. 400H (RIP) has a very different look to Portra 400. And then there's Kodachrome and Ektar. The recent effort at reverse-engineering Kodachrome was eventually funded by a company that wanted to produce an authentic LUT to turn a photo into a "kodachrome" photo. The LUT is available, the developing process isn't.

Over in digital land, I've always felt that Fuji sensors leaned a bit towards green, Olympus (a very long time ago) tended to red/orange, and Canon actually does a fair job of being neutral (although, you certainly can fix that in DPP4......).

One idea that I want to experiment with a bit is the idea of producing my own LUT files that I can use to do inversions for each film type.

 

[grat]

That makes no sense to me. Why do you need a LUT to make something look like... itself? When you invert Portra 160 you will get warm shadows, typical Portra, without any LUTs. And as I shown above, Kodak Gold 200 looks warm without any extra effort. If anything, you'll need LUTs to "normalize" different emulsions for maximum life-like accuracy, or to mimic wet printing on certain papers (Negmaster does that).

Load negative scan, apply LUT. Instant inversion with curves and histogram set correctly.

Not that it's particularly different from using Darktable, and copy/pasting the recipe for inversion.

 

[4season]

For what it's worth, Lomography's "Digitaliza" is a relatively inexpensive way to hold film somewhat flat without the use of glass:
https://shop.lomography.com/en/accessories/film-scanners/digitaliza-120-scanning-mask

As for cameras, if yours offers a "pixel shift" type feature, that can be an inexpensive way to get a whole lot of pixels at no extra charge. The feature seems to be overlooked because it's only usable for stationary subjects, but that's no limitation here. And in the case of my Olympus Pen-F, the feature can generate 80 megapixel raw files: Not too shabby for such a compact camera, I think.

 

[warden]

For what it's worth, Lomography's "Digitaliza" is a relatively inexpensive way to hold film somewhat flat without the use of glass:
https://shop.lomography.com/en/accessories/film-scanners/digitaliza-120-scanning-mask

As for cameras, if yours offers a "pixel shift" type feature, that can be an inexpensive way to get a whole lot of pixels at no extra charge. The feature seems to be overlooked because it's only usable for stationary subjects, but that's no limitation here. And in the case of my Olympus Pen-F, the feature can generate 80 megapixel raw files: Not too shabby for such a compact camera, I think.

I hadn't seen that Lomo holder before. That's cool.

 

[grat]

I've looked at the lomo holder-- it's an interesting idea. At the moment, I like the V800 series holders-- except they're rather large for my eventual scanning project. May copy some elements of the Lomo setup when I get my 3D printer back online.

As for pixel-shift, see previous discussion about lens resolution and film resolution-- I'm not sure there's 80MP of data in a 36x24mm 400 ISO frame of film.

 

[DonW]

Canon 5D MkII's are going cheap these days. Great way to step up to FF for scanning.

 

[grat]

Canon 5D MkII's are going cheap these days. Great way to step up to FF for scanning.

I lose 11MP, live view, focus bracketing, nearly a full stop of DR, and gain a bit of noise.

No thanks.

 

[Lachlan Young]

I'm not sure there's 80MP of data in a 36x24mm 400 ISO frame of film.

There's more than that, but it's not necessarily image content, rather it's film information (granularity etc) - and the transition between these is at once both rather fluid and vitally important to transmitting the totality of the image. The problem is that most scanning approaches have an historical hinterland more closely tied to image reproduction in pre-press applications - and emerged in a world in which bigger than 13x prints off 135 (let alone mural prints) were relatively rare. The issue is how much resolution you need to deliver sufficient MTFsystem to adequately represent the characteristics of the emulsion losslessly (ie no added artefacting/ exaggeration of granularity or noise) in a realistically scalable manner that doesn't increase visual granularity beyond how it appears when prints are made by direct optical means to the same size. There are no free lunches or easy answers.

 

[Lachlan Young]

will not convince me that I need a larger than 5000x3400px scan from a 35mm ISO100-400 negative.

For most uses apart from big exhibition prints, a full bit depth 3000-4000px on the short side of any neg is likely plenty, if the imaging system is inherently delivering adequate MTF & very low noise. The problems start when people fixate on resolution over all the far more important variables.

 

[Adrian Bacon]

For most uses apart from big exhibition prints, a full bit depth 3000-4000px on the short side of any neg is likely plenty, if the imaging system is inherently delivering adequate MTF & very low noise. The problems start when people fixate on resolution over all the far more important variables.

this is true, but at the same time, resolution is pretty important up until you hit that “good enough” point, and in all honesty, being someone who started scanning with an 18MP camera, then moved to a 24MP camera, then a 32MP camera, 24MP for 35mm film was so far into “good enough” territory it wasn’t even funny. 18MP was pretty good and easily rendered a huge amount of film grain if you had a good macro lens and enough light to shoot with a decent DOF and could keep the film reasonably flat. It’s enough resolution that I can see resolution differences in lenses attached to the film camera that was used to expose the film. Literally. And that was before I had the Macro lens that I’m using now, which renders way more resolution onto the sensor than before.

there’s nothing wrong with being a purist about a bunch of this stuff, but at the same time, practical reality rules the day, and the practical reality is a 16+MP DSLR/Mirrorless camera that is at least 12 bits raw ,(14 is better), and has a good macro lens, and a good way to hold the film flat, and a good light source is a pretty solid “good enough” for a really large number of use cases for negative film. Anything else short of simply adding a higher resolution body, or higher resolution macro lens is just adding operational complexity for not nearly enough bang, at least for 35mm film.

 

[warden]

this is true, but at the same time, resolution is pretty important up until you hit that “good enough” point, and in all honesty, being someone who started scanning with an 18MP camera, then moved to a 24MP camera, then a 32MP camera, 24MP for 35mm film was so far into “good enough” territory it wasn’t even funny. 18MP was pretty good and easily rendered a huge amount of film grain if you had a good macro lens and enough light to shoot with a decent DOF and could keep the film reasonably flat. It’s enough resolution that I can see resolution differences in lenses attached to the film camera that was used to expose the film. Literally. And that was before I had the Macro lens that I’m using now, which renders way more resolution onto the sensor than before.

there’s nothing wrong with being a purist about a bunch of this stuff, but at the same time, practical reality rules the day, and the practical reality is a 16+MP DSLR/Mirrorless camera that is at least 12 bits raw ,(14 is better), and has a good macro lens, and a good way to hold the film flat, and a good light source is a pretty solid “good enough” for a really large number of use cases for negative film. Anything else short of simply adding a higher resolution body, or higher resolution macro lens is just adding operational complexity for not nearly enough bang, at least for 35mm film.

Thanks for your participation in these DSLR scanning threads Adrian. You have me reconsidering my initial opinion that I need a full frame DSLR to digitize negatives.

 

[Adrian Bacon]

Thanks for your participation in these DSLR scanning threads Adrian. You have me reconsidering my initial opinion that I need a full frame DSLR to digitize negatives.

you're welcome. I’m happy to relay my experience, but at the end of the day, you’re the only one who can decide what is best for you.

all that being said, I’ve said it before, I’ll say it again: if you already have a camera and lens, just get the additional parts you need to start and just start. There’s enough other practical realities to work out that will affect the quality of your output that the total resolution of your camera and lens is way less of an issue until those other things are sorted, and even then, you may find that however much resolution you have, it more than plenty for how you’re using the scans.

I do get the desire for maximum res. We all want as much fidelity as possible, but at the same time, the bar to good enough is actually quite a bit lower than most of us are willing to admit.

 

[MattKing]

The reason I would like to have a full frame digital camera for digitization is that I already have a full frame 35mm slide film copying setup, including an appropriate full frame macro lens, and there really is no practical way to use it with a smaller than full frame sensor.

 

[Lachlan Young]

there’s nothing wrong with being a purist about a bunch of this stuff, but at the same time, practical reality rules the day, and the practical reality is a 16+MP DSLR/Mirrorless camera that is at least 12 bits raw ,(14 is better), and has a good macro lens, and a good way to hold the film flat, and a good light source is a pretty solid “good enough” for a really large number of use cases for negative film. Anything else short of simply adding a higher resolution body, or higher resolution macro lens is just adding operational complexity for not nearly enough bang, at least for 35mm film.

Realistically 16-24mp is good enough for most applications that end users need (all the way up to LF too) - and most of them aren't making 20x24's, let alone 24x36's or 40x60's from 135 on any sort of regular basis. The problem is that they need a VW, want a Ferrari, but all too often end up with a Lada...

 

[wiltw]

Thanks for your participation in these DSLR scanning threads Adrian. You have me reconsidering my initial opinion that I need a full frame DSLR to digitize negatives.

One reason to use FF...back in the days of film, there are a significant number of accessories made to precisely position a 135 slide afixed to the camera body with a specific length of tube so that it could be duplicated precisely at 1:1 onto a duplicating camera.

I have not checked to see if today you can find a product that holds slides precisely afixed to a APS-C dSLR for 1:1.6 reproduction of FF film slide

 

[wiltw]

Realistically 16-24mp is good enough for most applications that end users need (all the way up to LF too) - and most of them aren't making 20x24's, let alone 24x36's or 40x60's from 135 on any sort of regular basis. The problem is that they need a VW, want a Ferrari, but all too often end up with a Lada...

And in fact few folks would exceed 16" x 20" print from 135 for the simple reason of 16.9x mangification of grain!