Canon 90D/M6II + Sigma 70mm Macro ART Lens Resolution

Adrian Bacon · Aug 13, 2021

For reference and for discussion if people wish to: https://en.wikipedia.org/wiki/1951_USAF_resolution_test_chart

This is scanning 135 format film frame size, Black and White (color will be lower resolution due to the bayer array) using a Canon 90D, 1/200 second shutter, ISO 100, xenon strobe light source, and Sigma's newest 70mm Macro ART lens, at the lens highest resolving aperture setting, which is f/5.6. I've taken this exact frame from f/4 to f/22 in 1/3 stop increments and f/5.6 is easily the highest resolving aperture. Even a third of a stop in either direction is visually less resolution.

By my eye, Group 6, Element 4 is resolved, which is 90.5 line pairs per mm. You need 2 pixels per line pair, so 181 pixels per mm, multiplied by 25.4 mm per inch gives us ~4600dpi resolution. If you prefer to call Group 6 Element 3, then it's ~4100dpi.

Either way, for 35mm black and white film, you'll be fairly hard pressed to get more resolution with any other scanning system.

Lachlan Young · Aug 13, 2021

4000-5000ppi seems to be about the mechanical/ optical limit for a lot of camera scan systems without much more heavy duty engineering solutions. If you do get time to do the analysis, finding out the 50% and 10% MTFs of your system would be interesting - I've found that high contrast resolution isn't a great means of comparison between different scanning systems.

Adrian Bacon · Aug 13, 2021

Lachlan Young said:
4000-5000ppi seems to be about the mechanical/ optical limit for a lot of camera scan systems without much more heavy duty engineering solutions. If you do get time to do the analysis, finding out the 50% and 10% MTFs of your system would be interesting - I've found that high contrast resolution isn't a great means of comparison between different scanning systems.

digital is not film. Digital sensors have a contrast response of 100% all the way up to the sampling Nyquist. The sensor in my system is 6900x4600 pixels and I’m scanning an area that vertically is from sprocket holes on the top to sprocket holes on the bottom, which means my 4600 pixels is spanning just over the 24mm, close enough to 25.4mm, and I’m resolving 4600 dpi…. Which means… there is no 50% or 10% TOC. Well there is, but they’re probably 4625dpi and 4650dpi.

this also means that the Sigma art lens resolves a lot more resolution than the sensor, or we’d be seeing the MTF of the lens as it rolled off, and apparently it doesn’t. I’d need a higher pixel density sensor, and nobody makes a sensor that has more pixels per square mm of sensor. If they did and it was even remotely cost effective, that’s what I’d be using.

Oren Grad · Aug 13, 2021

Last year I replaced my Nikon 5000 and 9000 scanners with a Sony A7RIV plus the 70mm Sigma Art Macro. The A7RIV was the first camera I've tested that reliably produces superior results to the Nikon scanners, with the performance criterion being faithful rendering of the character of the grain in my Tri-X negatives. The Nikons produce obvious grain aliasing, the A7RIV does noticeably better. The EOS 5Ds R that I tested previously didn't - between the Bayer filter and the limitations of its live view focusing I couldn't improve on what the Nikon scanners were giving me. The sensor in the EOS 90D / M6II (used as a single-shot device - all sorts of things are possible if one is willing to stitch) doesn't have enough pixels to be competitive on this performance criterion.

The Sigma lens is indeed optically excellent. It focuses natively to 1:1 with full-frame cameras, which makes it especially convenient for copying 35mm film. And it's a great value at the price.

Adrian Bacon · Aug 13, 2021

Oren Grad said:
Last year I replaced my Nikon 5000 and 9000 scanners with a Sony A7RIV plus the 70mm Sigma Art Macro. The A7RIV was the first camera I've tested that reliably produces superior results to the Nikon scanners, with the performance criterion being faithful rendering of the character of the grain in my Tri-X negatives. The Nikons produce obvious grain aliasing, the A7RIV does noticeably better. The EOS 5Ds R that I tested previously didn't - between the Bayer filter and the limitations of its live view focusing I couldn't improve on what the Nikon scanners were giving me. The sensor in the EOS 90D / M6II doesn't have enough pixels to be competitive on this performance criterion.

The Sigma lens is indeed optically excellent. It focuses natively to 1:1 with full-frame cameras, which makes it especially convenient for copying 35mm film. And it's a great value at the price.

I’d check the corner performance if you’re using a full frame sensor to scan 35mm film. You’ll probably find that even though you have the pixels, it’s not as good due to the lens, alignment issues, and field curvature. A full frame sensor is great if scanning 120 or larger film as you’ll get more total resolution, but for 35mm, an APS-C sensor is the sweet spot that gives best all around performance.

of course, people should use what they feel is best for them, and they’re the only ones who can really decide what is best for them. I posted my results as a point of reference.

Oren Grad · Aug 13, 2021

Adrian Bacon said:
I’d check the corner performance if you’re using a full frame sensor to scan 35mm film. You’ll probably find that even though you have the pixels, it’s not as good due to the lens, alignment issues, and field curvature.

I have. The lens delivers excellent performance across the field.

I expect that any losses in the far corners will be less than the price paid by having only half the pixels. But of course it would be good to do a controlled comparison to verify that. Perhaps I'm wrong!

Hmmm... would you be willing to participate in such a comparison? Do you have a Tri-X negative that you've copied with your 90D? I'd be happy to copy it in my setup as well and swap files with you. Or I could dig up a negative here that you could copy in your system. Probably we'd both learn something, and the comparison might also be helpful for others trying to find the right cost/benefit tradeoff for their needs.

Tom Kershaw · Aug 13, 2021

Oren Grad said:
I have. The lens delivers excellent performance across the field.

Are you using the Sony camera for photography as well as film copying work?

Oren Grad · Aug 13, 2021

Tom Kershaw said:
Are you using the Sony camera for photography as well as film copying work?

I have tried it that way, don't especially care for it as a general-purpose camera. But in keeping with the ground rules here, if you'd like to discuss further we should do it in the appropriate digital subforum.

Adrian Bacon · Aug 13, 2021

Old Gregg said:
How do you get to lpmm numbers? The target appears to be sold as a 2x2" square, then you need to frame it just right? Is there a standardized 35mm pre-exposed target similar to control strips?

I don't understand the question. The linked wikipedia article converts the group/elements to lp/mm if that's what you mean.

I suppose I could contact print my target onto film, but then we'd just be measuring the film resolution.

Adrian Bacon · Aug 13, 2021

Oren Grad said:
I expect that any losses in the far corners will be less than the price paid by having only half the pixels. But of course it would be good to do a controlled comparison to verify that. Perhaps I'm wrong!

Hmmm... would you be willing to participate in such a comparison? Do you have a Tri-X negative that you've copied with your 90D? I'd be happy to copy it in my setup as well and swap files with you. Or I could dig up a negative here that you could copy in your system. Probably we'd both learn something, and the comparison might also be helpful for others trying to find the right cost/benefit tradeoff for their needs.

Sure. I'm not really a tri-x shooter, so the only tri-x images I have are customer images and I can't use those as they're not mine. My website (link in bio) has a mailing address for sending film in, if you want to dig up a roll that you feel would be a good candidate for comparison and mail it in, I'd be happy to put it in the queue and run it through then share results. Should be interesting.

Adrian Bacon · Aug 13, 2021

Old Gregg said:
I was asking if you're shooting the target with the macro lens directly, or you're shooting a negative with the target on it. The results could be different for macro lenses as their resolution varies with distance, and field curvature can become a factor as well. Not arguing with your results, just thinking out loud.

The target directly. The distance is the same as if it was a negative.

Oren Grad · Aug 13, 2021

Adrian Bacon said:
I'm not really a tri-x shooter, so the only tri-x images I have are customer images and I can't use those as they're not mine.

Understood, no worries. I'll come up with something here. I may end up shooting a test target set up for this purpose - will think a bit about how to make this as informative as possible. Stay tuned...

Adrian Bacon · Aug 13, 2021

Oren Grad said:
Understood, no worries. I'll come up with something here. I may end up shooting a test target set up for this purpose - will think a bit about how to make this as informative as possible. Stay tuned...

OK. If you send anything in, put a note in with it referencing this thread so I can identify it and set it aside so you don't get billed for digitizing the roll. If possible, uncut is better in terms of time it takes to handle and scan it in, but cut and sleeved is fine as well. Also, I can control the applied contrast, so if you know the straight-line gamma that you develop to, put that in with the note and I'll use that so both systems should produce roughly the same scanned contrast.

Oren Grad · Aug 13, 2021

Adrian Bacon said:
OK. If you send anything in, put a note in with it referencing this thread so I can identify it and set it aside so you don't get billed for digitizing the roll. If possible, uncut is better in terms of time it takes to handle and scan it in, but cut and sleeved is fine as well. Also, I can control the applied contrast, so if you know the straight-line gamma that you develop to, put that in with the note and I'll use that so both systems should produce roughly the same scanned contrast.

Duly noted, thanks. I'll contact you privately with details once I settle on a plan, to make sure it seems reasonable to you. It may be a few days.

Adrian Bacon · Aug 13, 2021

Old Gregg said:
Related anecdote which re-iterates Adrian's point: I have played with 61MP Sony a7R IV paired with Sony FE 90mm f/2.8 macro G OSS lens trying to see how it compares to my more pedestrian 30MP Canon 5D Mk4. I could definitely see a slight resolution boost in the middle (a tiny patch, really) but there was no improvement outside of it and the corners were actually much worse. This is a 5-star rated lens which turns your 60MP camera into a 20MP camera for about a thousand dollars.

Not that I'm looking to start a resolution discussion/argument/war or anything, but in my experience, ~24-30 MP for a 35mm frame of film is where things start to become very sensitive to other factors that have a bigger impact on the effective captured resolution than the raw sensor/lens performance, and it really at that point becomes a balancing act of getting the best overall compromise. Just because you're capturing more pixels doesn't mean you're actually capturing more detail. I've totally been foo-barred by things as simple as film that won't stay flat, vibration during the capture, etc. My current system is the result of an evolution over time of identifying and chasing down sources of detail loss and either eliminating or reducing them as much as realistically possible to get to a place where I have very good performance, and can very productively scan film with a package that isn't particularly cheap, but gives outstanding bang for the buck spent.

albireo · Aug 14, 2021

Oren Grad said:
Last year I replaced my Nikon 5000 and 9000 scanners with a Sony A7RIV plus the 70mm Sigma Art Macro. The A7RIV was the first camera I've tested that reliably produces superior results to the Nikon scanners, with the performance criterion being faithful rendering of the character of the grain in my Tri-X negatives. The Nikons produce obvious grain aliasing, the A7RIV does noticeably better.

There is absolutely no grain aliasing produced by a well maintained and correctly used Coolscan 8000 or 9000. How many units did you test in support of your statement above? If the above is derived from experience with your own unit, I'd suggest it might need a CLA or recalibration.

In the USA, Frank A. Phillips does extraordinary work on these machines. My 8000 was tuned up by him and produces better results than Bayer or Xtrans sensor equipped DSLR setups (the variability is of course considerable, and will depend on the final user rather than on a team of Nikon engineeers) at a fraction of the price.

Here, my definition of "better" is of course rather personal and to me means much more than peak resolution at the macro lens sweet spot.

Andrew Patteson · Aug 14, 2021

I'll just add on a more qualitative note that Adrian knows what he is talking about -- I am a customer of his lab and have been very happy with the scans. Better results and equal or better price than the Frontier/Noritsu-based services I have tried. Being able to receive raw files is great, too.

Adrian Bacon · Aug 14, 2021

For those that wish to zoom into 1:1 on the 90D and stitch their scans together: an eye watering ~7300-8200DPI for BW. Basically, whatever resolution your sensor is, this lens is the one to have, at least in the center. The jpeg compression here is killing it a little, but rest assured, group 7 Element 2 is clearly resolved, and I'd be willing to call Element 3.

I don't personally subscribe to stitching simply because it is very time consuming and fraught with all kinds of other problems that will easily wipe your resolution right off the table, but, for those special frames, if you have this lens and a 90D/M6II, you can go there if you really want to.

grat · Aug 14, 2021

You're making me consider putting my 100mm f/2.8 macro USM up for sale. You're a bad, bad person.

Adrian Bacon · Aug 14, 2021

grat said:
You're making me consider putting my 100mm f/2.8 macro USM up for sale. You're a bad, bad person.

Sorry. You can probably sell it and pick this one up for about the same price, so a near break even, or maybe only a couple hundred difference. I've not looked at what the 100 macro USM is going for lately, so I could be wrong. The 100mm macro is good, but isn't totally awesome in the center. It just maintains the same level over almost the entire frame, so on a full frame camera it's quite good because it's really consistent over the entire frame, but if you're shooting a high resolution APS-C body, or a much higher res full frame body like the 5Ds or adapted to the R5, the sigma is a better lens.

grat · Aug 16, 2021

Interesting-- they reviewed the "L" version of the Canon EF 100mm f/2.8, and complained of color fringing-- a complaint I have never, ever heard leveled at the non-L version (and I just looked again to see if I'd missed one). On my 90D, the pixel density is definitely pushing the boundaries of what is now a pretty old (for a modern camera) lens.

Unfortunately for sellers, with the RF lenses coming out, the EF lenses have take a bit of a hit on selling prices.

Adrian Bacon · Aug 16, 2021

grat said:
Interesting-- they reviewed the "L" version of the Canon EF 100mm f/2.8, and complained of color fringing-- a complaint I have never, ever heard leveled at the non-L version (and I just looked again to see if I'd missed one). On my 90D, the pixel density is definitely pushing the boundaries of what is now a pretty old (for a modern camera) lens.

Unfortunately for sellers, with the RF lenses coming out, the EF lenses have take a bit of a hit on selling prices.

I concur. Interesting read, though, pretty much what I expected. I'm curious to see what the RF 100mm Macro will do. I've got an EOS R5 on hand that I'm evaluating for doing extra large scans with and would like to potentially migrate from the 90D to RF mount at some point. My hope is that Canon releases an APS-C sensor RF camera that is 7680x5120 (39MP) and releases a full frame camera that has 12288x8192 for ~100MP. That would be pretty sick for 120 and 4x5 film.

7680x5120 is the next logical step for APS-C as it's a nice ~7MP jump up from the 90D, and is a very nice and evenly divisible number for the video crowd that wants to do UHD. Having quad UHD as the sensor source scaled down to UHD will make for a nice crispy image. The same goes for the full frame, it's a nice easy to scale (per DCP frame sizes) step up and gives a big doubling of MP over the standard R5, which is the same as what they did when they released the 5Ds. At the time, the 5DII was 22MP, so the jump to 50MP seemed huge. If Canon knows what's good for them, from here on out, they'll keep their frame sizes nice even multiples of either UHD or DCP video frame sizes to keep scaling the video simpler for the video shooters as it's become pretty obvious that good hybrid cameras are the future to sales. There's still a place for stills, but more and more video is an integral part of capturing imagery, and if you're going to drop a few thousand dollars on a camera body, it better be pretty capable of both stills and video.

Adrian Bacon · Aug 17, 2021

Old Gregg said:
@Adrian Bacon why not switch to Sony A7R IV? It's got an amazing implementation of 4-shot pixel shift (adds zero overhead when shooting tethered - you get a merged RAW automatically) and its pixel density is right there where you need it. Mount your Sigma glass in a native mount and you've got enough pixels to crop the extreme corners out. I was blown away, just didn't have a matching lens to take advantage of it...

Ugh… multiple reasons…. For one, every single Sony camera I’ve used (and believe me, I’ve rented and tried quite a few), I feel like I’m fighting the camera just to use it. It’s exhausting. Sony and many Sony users suffer from what I call “featuritis syndrome”. The act of packing more and more features and “gee whiz” gizmo stuff into their cameras in an attempt to steal market share from Canon, most of which isn’t fully baked, nor really needed, which then just gets in the way of trying to take pictures. I can’t tell you how many times I find myself wishing I could turn off something the camera is doing because it’s getting in the way. Feel free to call me old fashioned, but that’s just how I operate. It’s not just Sony either. I just feel less offended when using Canons cameras. I’m at a place where using an old manual camera that doesn’t even have a battery is such a pleasing experience. I can just zone in and get in that creative place where I just see stuff and capture it. It’s magical, at least to me, and almost all of my favorite pictures are when I’m there in that place.

number two, Sony’s sensor color response…. Color me extremely meh… on top of that, that means I have to go and generate new scanning profiles for Sony. Even more Meh… more work than I want to do, and I’m busy enough I’d have a really hard time getting the time to do it.

number three, I’d have to build in support for reading Sony’s raw files. Extra super awesome meh. I have support for Canon’s CR2 and CR3 files and canon has so far been extremely consistent in their sensor color response from one camera model to the next which has made switching camera bodies pretty straightforward with a few tweaks here and there.

number four: I don’t actually need the resolution. In fact, the number one thing I get in terms of feedback from customers is “Your files are so huge!!! Is there any way to make them smaller?” Yes, yes there is. This is why I now offer scanning in multiple sizes. My most popular size is the medium size, which is a quarter of the raw sensor resolution, but at 8MP still spectacular when displayed.

also, I’m not going to jump ship over something like that. Canon isn’t going to let Sony sit at the top of the resolution chain for long. Let’s not forget, Canon historically has been the first to reach resolution high water marks, and they’ll do it again.

grat · Aug 17, 2021

Old Gregg said:
I hear you regarding usability. Speaking of "sony color", isn't that 100% in software though?

Not 100%, no. The choice of filters used in the Bayer array (or other) will influence the final image-- Consider tri-negative color photography where you use three negatives, each with a different color filter. The typical tri-color sets are 25/58/47 and 29/61/47B, which will produce different "color" images.

Site like DPReview rave over the color saturation of the Sony cameras-- But I've noticed commercial sports / wildlife photographers tend to really like the neutrality of the Canon sensors. Even in the age of digital, you go to a major sporting event, and the sidelines are heavy with large, white lenses (usually Canon L series). Be interesting to see how fast that changes now that the RF lenses are out.

Adrian Bacon · Aug 17, 2021

Old Gregg said:
Speaking of "sony color", isn't that 100% in software though?

Nope. Digital camera sensors do in fact have a native white balance that is a combination of the spectral characteristics of the CFA filter and native sensitivity of sensel underneath it along with whatever baked in gain they put in relative to the other color channels that you can't control. It's pretty easy to work out what the native WB is. Just shoot a series of photos where you manually set the white balance up and down the range in 100K increments then look at the white balance multipliers in the raw file. The kelvin setting that nets the red and blue multipliers closest to each other is the native white balance for that camera/sensor. The other thing that affects things is what the actual multiplier ratio is between the red and blue channels and the green channel. People always assume that looking at a raw bayer array output looks green because there's twice as many green pixels. Nope. It's because the green channel values are brighter. If you apply the red and blue WB multipliers to the red and blue channels, the green tint goes away, even when looking at the raw bayer array samples. Then there's how much spectral overlap there is between each of the color channels when the camera sensor is getting hit with full spectrum light. That overlap affects things like the native saturation of color before any color processing happens.

The combination of those things has a huge impact on the color processing after capture. Canon primarily has two native white balance centers between their various models: ~5200K and ~4500K. If the camera is more likely to be shooting in mixed lighting (i.e. indoors and outdoors), the native WB tends to be ~4500K. If it's more likely to be used outdoors or with studio strobes, then it's ~5200K (this is why when you set their white balance to daylight, it's 5200K, not 5500K that you see in LR). Sony on the other hand tends to be considerably warmer in their native sensor WB (closer to 3200-4000K, though I've not inspected every camera I've tried), and in general has a significantly different color response than Canon.

Old Gregg said:
if you're using something like Lightroom, it's the Adobe color you are be looking at

My code reads the raw sensor values and generates something that you look at in Adobe LR. When you pull my DNG files in, you're pulling in image data that has been conformed from the raw values containing the negative image to a positive image that LR can do something with. It takes the emulsion native color response (which you can look at in the the spec sheet for each film) as seen through the camera's native sensor color response with the light source you're using and turns it into a color positive image that looks correct (for the most part) when in LR after the fact. Just dropping another sensor in there with a dramatically different color response without doing anything else totally hoses the colors.

Canon 90D/M6II + Sigma 70mm Macro ART Lens Resolution

Privacy & Transparency

Privacy & Transparency