Wetpixel Live: Super Resolution for UW Photographers

A possible game-changer?

Adobe recently released their new Super Resolution function in Photoshop. Based on machine learning this allows for significant and very accurate upscaling of images. @Alex_Mustard and @adamhanlon discuss how this may affect underwater photographers' choices of camera and allow for a whole new criteria when assessing camera performance.

I just recently sold some (land) prints and I tried to use this feature it was disappointing. I got better results with preserve details 2.0

This was a smaller upscale from 5184 to 6000 and yet the new feature did not do that well and added quite a few artefacts.

In general however you really don't need more than 20-24 megapixels as you can print fine at 200 DPI and printing lab upscaling is very impressive

I print 30" wide and most people do not have room for larger frames in their houses unless they are the kind of people that would not typically buy a wildlife image

There are conflicting reports on super resolution but the general message of this video aka you are good with a 20-24 megapixel camera remains valid

I watched the segment with great interest, a thoughtful discussion by Adam and Alex as usual.

I tried it out on a couple of files from 2012, taken with my trusty old D2X. 12.2 megapixel files which up-rezzed to 40 something megapixels. They looked really good and I saw no real artefacts. 

Alex discussed how this was a game changer, in that a lower resolution platform, such as a D300 could become relevant again. This is an exciting concept for sure, but it ignores the fact that resolution is not the only criterion for "progress" in camera design.

I currently own a D850, previously D2X, D3, D3X, D800, D810 etc and the march of progress has been undeniable. Not only with respect to resolution, but more importantly, autofocus, and dynamic range of the sensor. These attributes are not part of the Super Resolution algorithm, and are probably much more important that raw resolution. Low light performance and diffraction may be other matters though.

I think for photographers just starting out, this may represent a great way into the hobby; buying older gear and housing for a fraction of the price of 'bleeding edge' gear. It might also slow down the arms race to better and newer gear each time a new camera is introduced, if the main improvement is strictly resolution.

However, I feel that the D850 is currently at the top of the heap all round. It is hard to imagine much more improvement coming the way of DSLR technology. I am happy to stick with what I've got, and likely will NOT replace this camera in the future. It really seems to be a near perfect tool for the endeavour.

Mirrorless may be a completely different matter:)

I'm not selling my D850 and buying a D300 either! Actually the smaller photosites on cropped sensors are probably not where the real advantage lies. As Alex points out, the larger photosites on older full frame  (say 5D Mark II/Nikon D3) are probably where this technology could have big impact. 

As we mentioned though, newer high resolution cameras are harder to shoot, more particular about optics and more prone to emphasizing any technique errors...

Totally agree, Adam.

I did toy with the idea of taking my D3S out of mothballs and finding a nice Subal housing for it.

Then I gave my head a shake...

But it is nice to think that with the march of computer technology, we might be able to remain satisfied with the platforms we have for much, much longer. I know for sure this would make my missus happier:))

ian

1 hour ago, ianmarsh said:

But it is nice to think that with the march of computer technology, we might be able to remain satisfied with the platforms we have for much, much longer

In our dreams, Ian....

Adobe ACR does not do a particularly good job in terms of image sharpness compared to other raw converters

The engine has been updated a lot since you reprocessed your shots from 10 years ago and the improvement are likely to be connected to the new processing logic not so much to super resolution

There are plenty of examples out there where super resolution does not provide any benefit compared to other software and today there are many solutions for scaling including Topaz products for example or a combination of a better RAW processor with photoshop enhance details

In addition super resolution only works on RAW not on TIFF files which is another limitation

It works for RAW, TIFF and JPG. To get it to work, you need to open them from Bridge. I understand that a future release of Lightroom will also allow the opening of files into Super Resolution too.

In terms of use, my testing of it thus far is that it is far, far better than Nik Blow Up and Gigapixel AI, particularly with low res images. Depending on the original file's quality, I have found it difficult to distinguish between the original and the up-res'd version and in "blind" testing have been unable to do sometimes. By comparison, artIfacting and inferior pixel level interpolation of both the Nik and Topaz products is immediately visible.

Predictable, it works better on some files. I think it seems to cope well with some optical/technical flaws, but exaggerates others. Any overexposure seems to be nasty, but shadows seem to retain detail. 

Testing is continuing

53 minutes ago, adamhanlon said:

It works for RAW, TIFF and JPG. To get it to work, you need to open them from Bridge. I understand that a future release of Lightroom will also allow the opening of files into Super Resolution too.

In terms of use, my testing of it thus far is that it is far, far better than Nik Blow Up and Gigapixel AI, particularly with low res images. Depending on the original file's quality, I have found it difficult to distinguish between the original and the up-res'd version and in "blind" testing have been unable to do sometimes. By comparison, artIfacting and inferior pixel level interpolation of both the Nik and Topaz products is immediately visible.

Predictable, it works better on some files. I think it seems to cope well with some optical/technical flaws, but exaggerates others. Any overexposure seems to be nasty, but shadows seem to retain detail. 

Testing is continuing

 

Apologies I was talking about the full shebang with RAW Details which is what bring the most benefit, also suggesting their own RAW processing is not that good if there are details that can be revealed?

If you base your comparison on the  ACR RAW processing am sure you will see benefits

However when I say it does not work that well I am comparing with a TIFF produced by DxO that is way sharper than any image lightroom can produce.

So the path DxO or RawTherapee or other RAW to TIFF and then a sharpener works better for me than processing a RAW in Adobe products and this feature is not really making me change the workflow especially once I factor in all the issues with color rendering and exposure that adobe introduces when it normalises a camera