With Super Resolution, do we need 45+ MP cameras?

We don't seem to have discussed this much on here, but I think it is worth talking about Adobe Super Resolution and how we can use it.

This potentially offers the option of increasing the resolution of files by 4 times, so a 12MP file, becomes a 48MP one for example. Adobe uses AI and machine learning to make it work, and my experience is that their algorithm seems to work well, albeit with some provisos.

I have found that it works really well with files that are optically good and it really enhances any optical flaws. So images shot with good lens/port combinations work really well. I have been using it with files from my D2Xs, and it is really giving new life to them. I think the inherent optical advantages of cropped sensor SLRs behind ports really plays into this. 

For example, you can take a D850 and WACP and produce excellent quality wide angle images at 46MP. Arguably, you could do the same with the camera and a 16-35 f/4 with a 230mm dome, keeping the aperture at f/11 or above.

Or

You could take a D500, with a Nikon 10-24 f3.5-4.5 and a 230mm dome and run the images through Super Resolution to get an 80MP file (!) with similar or better quality. You will get a stop or two of aperture that will be usable and still get good corner sharpness that is as good or better than that with the D850..

Of course, I am using a Nikon analogy, but the asme could be applied with other camera brands. The Sony a7s III is a super exciting option with its big "un-noisey" pixels , albeit with limited (or expensive) lens options!

Another option would be to use legacy cameras, like the D2Xs, that are now incredibly good value, and when paired with the right optics, potentially capable of producing results that are on a par with newer full frame cameras.

Has anyone else been experimenting with Super Resolution? Would be super keen to compare results...

This is a great topic Adam!  I agree that computational tactics have a future, even if the purist SOOC crowd cringes . . . 

I’ve been playing around with the equivalent program from Topaz, Gigapixel AI (as well as their Sharpen and Denoise modules). So far it seems very promising.  I’ll try to produce a good example for comparison.  

While super resolution does provide an improvement in image quality, it does not compare to actual recorded pixels. The exception are in camera multishot modes that don’t work underwater in most circumstances. It closes the gap a little bit,  which in my opinion as a mft shooter isn’t that large anyway, but there is no denying that real 60mp files from a current camera hold more detail than a 24mp file pushed to even 80mp at least when looking at high magnification on the screen. I’ve not tried printing a super resolution file as I rarely print large enough for it to matter. I Would be Interested in experiences from those who printed a super resolution file big. 

I am mostly using SuperResolution or Topaz Gigapixel for images that I have cropped significantly from a D850 image. For example, I might take a landscape image and crop it to portrait and then upsize. Making 8x10 portrait crop drops from 8200 -> 4400 pixels wide. You lose almost half the pixels.

Below is an example of an image of a Giant Pacific Octopus I took last week at nearly point-blank range crammed into a crevice. There was no way practical way to turn the camera sideways in this position. I wanted to use this image in portrait mode for a cover shot so I cropped it 8x10. SuperResolution returns it to 8800 pixels wide.

I primarily do supermacro with a Sony a6500+ sony 90mm + Nauticam SMC-1 or CMC-2 and super resolution has halted my interest in upgrading to full frame. Even images with 50% + crops can be resuscitated for large prints. Frankly for Wide angle the Sony 10-18 is a better underwater lens option than what you can pick from Fullframe. 

It does not work so well with Nikon D1x files because of the asymmetrical pixels, ROTFL. Here is a screen shot from an early version of my "Salmon Year" slideshow in LR. You can see a few doubled images in the filmstrip where I did some supersizing (my preferred name; Adobe should call the button "supersize me", ROTFL). Note the highlighted image in the filmstrip and the one to its right (it is not narrow as the shot was taken looking straight down). I took this shot in 2005.

I added a second screen shot in grid view to make this clearer. The two other supersized pix that you can see in the grid were shot with the D2H (a 4 MP camera).

There is no doubt in my mind that Super Resolution works better with some files. 

My experience is the better the file, the better the process works. This seems to be somewhat independent of its original resolution, but more it’s optical and exposure properties. It really exaggerates any optical or technique flaws in the camera system or shooting technique.

I haven’t done lots of testing, but it also seems that sensor size is a factor too. My results are not great with m4/3 or compact sized sensor produced images. This may well be to do with the machine learning aspect, perhaps Adobe has not really concentrated on these types of image. If this is so, it may well improve with future updates.