Does Adobe's Super Resolution in ACR obviate the M10-R and M10 Mono line?

[dante]

Advertisement (gone after registration)

Serious question: if you've tried Adobe Super Resolution, has it affected your attitude toward buying an M10-R or M10 Monochrom?

I didn't think much of this but recently blew up both some M240 and M246 pictures using this new feature of ACR, which generates a double-linear-resolution DNG from your original DNG, so 96mp (these are absolutely massive files in color). Having looked at the output, especially with the Mono, I'm hard pressed to see how at 100% enlargement (now a 40" wide print at 300dpi), the quality is much degraded from the original 24mp shot. In fact, it's all but impossible to even tell an image was reprocessed at all. The implication for me is that my M cameras are now encroaching on the turf of my medium-format B/W gear (96mp is around the effective resolution of a scanned 6x9 TMY negative).

So if you got 96mp out of an M240 or M246, would you spend 8 grand for a new 40mp M body? I'm starting to think that my current bodies will be around for a good while.

Edited April 24, 2021 by dante

[Advertisement]

The Leica Q3 is available for pre-order at B&H Photo Video and Adorama

[mmradman]

I must try that, thanks.

I think digital photography is like Olympic Motto, Citius, Altius, Fortius, we always seem to strive for better and better.  In the early days sensor resolutions were low and bicubic transforms helped with large size prints, nowadays if you have 40Mp camera why not try to double its linear resolution because you can.

 

[frame-it]

38 minutes ago, dante said:

my M cameras are now encroaching on the turf of my medium-format B/W gear

40+mp on a 35mm sensor is nothing like 40+mp on a mini MF [x1d/GFX] or full MF [P1/Hasselblad H] Sensor

People forget the Sensor dimensions and only talk about the megapixels...strange

[Good To Be Retired]

I did some experimentation with Super Resolution when it arrived and determined that, at least for my personal needs, it replaces nothing.

While I have seen some tests where it appears to have made a real difference, I was unable to duplicate that result.

[dante]

30 minutes ago, frame-it said:

40+mp on a 35mm sensor is nothing like 40+mp on a mini MF [x1d/GFX] or full MF [P1/Hasselblad H] Sensor

People forget the Sensor dimensions and only talk about the megapixels...strange

I don't get this argument. An M240 has a slightly wider pixel pitch than a GFX50 or X1D, and an M10-R/A7R2 has a wider pixel pitch than a GFX100. For a given megapixel count, pixel size is the only thing medium-format gets (or really, got) you. The H6D at a cool $33k still has a slightly smaller pitch than an M240. You could argue some incremental quality loss in a doubled (really, quadrupled) 24mp sensor versus a 100mp of the same pitch, but for most people, it's not a $25,000 quality increase. 

[jaapv]

2 hours ago, dante said:

Serious question: if you've tried Adobe Super Resolution, has it affected your attitude toward buying an M10-R or M10 Monochrom?

I didn't think much of this but recently blew up both some M240 and M246 pictures using this new feature of ACR, which generates a double-linear-resolution DNG from your original DNG, so 96mp (these are absolutely massive files in color). Having looked at the output, especially with the Mono, I'm hard pressed to see how at 100% enlargement (now a 40" wide print at 300dpi), the quality is much degraded from the original 24mp shot. In fact, it's all but impossible to even tell an image was reprocessed at all. The implication for me is that my M cameras are now encroaching on the turf of my medium-format B/W gear (96mp is around the effective resolution of a scanned 6x9 TMY negative).

So if you got 96mp out of an M240 or M246, would you spend 8 grand for a new 40mp M body? I'm starting to think that my current bodies will be around for a good while.

Super resolution enhances detail  that is present in the raw data (which, given some fancy Photoshop techniques as described by Dan Margulis could be done without this automation) making the image appear more detailed. However, what it cannot do is render the extra fine detail that is produced by high-resolution sensors, as it is simply not there in the file. So no, it cannot replace a high-resolution sensor. And it can use a 40 MP file and give it the same boost again.

[Ray Vonn]

Advertisement (gone after registration)

It’s quite time consuming going through that process in PS, you wouldn’t want to be doing that on every image.

[dante]

On 4/24/2021 at 8:41 PM, jaapv said:

Super resolution enhances detail  that is present in the raw data (which, given some fancy Photoshop techniques as described by Dan Margulis could be done without this automation) making the image appear more detailed. However, what it cannot do is render the extra fine detail that is produced by high-resolution sensors, as it is simply not there in the file. So no, it cannot replace a high-resolution sensor. And it can use a 40 MP file and give it the same boost again.

No doubt that you capture tinier things with a higher-res sensor, given the same lens (assuming the lens resolves them... which is not a given with a lot of older glass, especially focused by old eyes). I think high MP sensors help sell expensive lenses (which have no point unless you can blackboard a big system resolution) and facilitate steep crops. Those sensor also help preserve a sufficient data volume with more severe forms of post-processing perspective correction. But the world has largely settled around 24mp mirrorless cameras, which suggests that statistically, that's acceptable for most uses. We're coming onto the eighth or ninth year for the Sony A7 line where the main camera still 24.2mp and includes models as late as July 2020 with 12.1mp(!) sensors. 

And certainly, a massive uptick in responsiveness or reliability in a camera can make an incidental upgrade compelling. For example, the M8/M9 architecture got to an 18mp resolution, but from the M240 on, the user experience improved considerably. More data was just a plus.

My questions about obviation have more to do with having sufficient resolution for any size output than they are about catching the tiniest cheek fuzz of a 21-year-old woman in a head shot. I'm not sure that the smallest meaningful aesthetic level of detail has really moved off the "texture of concrete" or a "non-aliased eyelash." Even in landscapes, I'm not sure that any picture has been made or broken by being able to resolve a telephone wire at 1000m vs 2000m. If there is an identifiable subject in a picture, it's hard to point to one that is defined primarily (or even appreciably) by the smallest thing a sensor or film grain can see. There is an "ooh and ah" factor to the microscopic, but that's for discussion sites.

In prints, the only place you are not discarding most of the data for an output, the real mark a camera needs to hit is to create enough data about sufficient levels of detail aesthetically and enlarge for a normal viewing distance without adding blur or artifacts. Ansel Adams believed that this might take as few as 1 megapixels, which is obviously low, but the point remains that the bigger the print, the greater the normal viewing distance. This is the theory upon which depth of field relies - circles of confusion. By the numbers, you need 120dpi at a 1.5m viewing distance. From an M9, that's a 43.3" (110cm) print, which is frankly enormous. Obviously with the M240/M10 you get a 50"/127cm output at that resolution.

My testing of Super Resolution was prompted by a project where I was making large prints (18"/46cm short side) for a client from 24mp M files and wondered what the resulting image files (300dpi) would look like in an unrealistic 100% view on screen. If they don't show artifacts at that size, unrealistic as it is, there is zero real-world chance that anyone would detect issues in the finished prints, no matter how close they got to them. And when the prints from this experiment came back yesterday, they stood up to the 40cm scrutiny test (that's about the expectation for an adult's eye to even focus). So no pixels, no artifacts on a close-up look in person, on a print. The ultimate installation location for these would not allow a human to approach closer than 1.5m, so rescaling was always overkill anyway.

I did read Margulis' article on unsharp masking and Super Resolution. I know what he wrote, but if his illustrations are indicative, his conclusions seem to assume a much higher tolerance for halos on high contrast edges. I didn't see the results as exactly equal, and the point of the machine learning is to cut down on that. Super Resolution does have some limitations with the Monochrom (you can occasionally see something weird, though you would need 200% magnification on screen to see it).

Dante

 

 

[01af]

Adobe's new Super Resolution feature yields incredibly good results, much better than plain good ol' bicubic upscaling in ACR or Photoshop ever did. Yet, an image from a sensor with a natively higher pixel count will be better still. So no, Super Resolution doesn't obviate high-megapixel sensors. Moreover, you can always apply Super Resolution to high-megapixel images, to get even bigger images ... whatever pupose it may serve.