Adobe super resoloution vs pixel shift

[James Stirling]

I appreciate that pixel shift has other advantages than just resolution ( colour , moire, noise etc ) but for a process that only takes a couple of seconds . The Adobe super resolution surprised me . I know it has been in ACR for a while but this is the first I compared it directly

RAW base ISO Dpreview sample standard and pixel shift . I opened the pixel shift in workspace then into photoshop for the side by side

Bearing in mind these 100% crops equate to a 35" wide print at the 300PPI it is not too bad at all.

100% comparisons from various parts of the scene





You can see the advantage of the pixel shift ( on the right ) best here in the smallest print, where it really does a better job. It is also cleaner .



And yes I am a long term persecuted tenant of Adobe software, where the cruel landlord raises prices on a whim and can be slow repairing something

--
Jim Stirling:
"To argue with a person who has renounced the use of reason, is like administering medicine to the dead." - Thomas Paine
Feel free to tinker with any photos I post

 

[MEDISN]

Interesting Jim, thanks for sharing. Now that Adobe puts Super Resolution (along with raw details and denoise) in LR, I find it a lot more useful (accessible?). Although the results are less persuasive than with your studio scene examples.

View attachment 9f79b869088941aea01b841028f365cc.jpg
PEN-F pixel shift

View attachment 162923218b81403587503d5d16560e4f.jpg
Adobe Super Resolution from the 20MP ORI

There seems to be sharpening of shot noise in the image giving a strange look to some areas. Perhaps I should use Denoise first followed by Super Res.

In any case, for times I can use HR or HHHR, I just assume do so. For times I can't, it's nice to know I may be able to get a little boost from the software side. They've come a long way, that's for sure.

 

[James Stirling]

Interesting Jim, thanks for sharing. Now that Adobe puts Super Resolution (along with raw details and denoise) in LR, I find it a lot more useful (accessible?). Although the results are less persuasive than with your studio scene examples.

A static studio scene with ideal lighting etc is a lot less challenging than the real world. A lot more components of the scene to deal with , even so the results are pretty decent. Like my face if you look too close all kinds of horrors await you

View attachment 9f79b869088941aea01b841028f365cc.jpg
PEN-F pixel shift

View attachment 162923218b81403587503d5d16560e4f.jpg
Adobe Super Resolution from the 20MP ORI

There seems to be sharpening of shot noise in the image giving a strange look to some areas. Perhaps I should use Denoise first followed by Super Res.

A good idea sometimes if you are going to be doing some generous editing a touch of NR can help, as the process can add some degree of noise

In any case, for times I can use HR or HHHR, I just assume do so. For times I can't, it's nice to know I may be able to get a little boost from the software side. They've come a long way, that's for sure.

I don't print at home any more but I will try a couple of large prints with an image from my OM-1 upscaled and pixel shift . I suspect even on a fairly large print that what is so easy to see on screen may not be so much of an issue. Prints are a great equaliser

--
Jim Stirling:
"To argue with a person who has renounced the use of reason, is like administering medicine to the dead." - Thomas Paine
Feel free to tinker with any photos I post

 

[knickerhawk]

Interesting Jim, thanks for sharing. Now that Adobe puts Super Resolution (along with raw details and denoise) in LR, I find it a lot more useful (accessible?). Although the results are less persuasive than with your studio scene examples.

In my experience, the benefits of Superresolution over resampling using the Preserve Details 2.0 option are mixed. Where Superresolution works, it tends to do a better job with improving fine detail acutance. However, it's also more prone to generating ugly and visible patches of what look like over-sharpened/over-saturated fine detail gone awry. It's pretty random and unpredictable, which means you have to carefully examine the whole image to see if these patches show up.

View attachment 9f79b869088941aea01b841028f365cc.jpg
PEN-F pixel shift

FYI, I'm still playing with using FFTs to reduce the motion and processing artifacting in pixel shift shots. Haven't decided if the improvements are easy enough to generate on a repeatable basis to merit an in-depth posting about it. The above pixel shift shot is full of "interesting" artifacts that would make it a good candidate for analysis if you want to forward the raw to me.

View attachment 162923218b81403587503d5d16560e4f.jpg
Adobe Super Resolution from the 20MP ORI

There seems to be sharpening of shot noise in the image giving a strange look to some areas. Perhaps I should use Denoise first followed by Super Res.

That hasn't worked out well for me. It's as if the AI-generated artifacts generated by the denoise pass are just compounded by artifacts added by the superresolution pass. I've only tried it on a few shots, so results may vary.

In any case, for times I can use HR or HHHR, I just assume do so. For times I can't, it's nice to know I may be able to get a little boost from the software side. They've come a long way, that's for sure.

 

[MEDISN]

Interesting Jim, thanks for sharing. Now that Adobe puts Super Resolution (along with raw details and denoise) in LR, I find it a lot more useful (accessible?). Although the results are less persuasive than with your studio scene examples.

In my experience, the benefits of Superresolution over resampling using the Preserve Details 2.0 option are mixed. Where Superresolution works, it tends to do a better job with improving fine detail acutance. However, it's also more prone to generating ugly and visible patches of what look like over-sharpened/over-saturated fine detail gone awry. It's pretty random and unpredictable, which means you have to carefully examine the whole image to see if these patches show up.

The "patches" are what I noticed in my example.

View attachment 9f79b869088941aea01b841028f365cc.jpg
PEN-F pixel shift

FYI, I'm still playing with using FFTs to reduce the motion and processing artifacting in pixel shift shots. Haven't decided if the improvements are easy enough to generate on a repeatable basis to merit an in-depth posting about it. The above pixel shift shot is full of "interesting" artifacts that would make it a good candidate for analysis if you want to forward the raw to me.

Oh yes, I added it to your Dropbox folder as a second PENF sample. Not only is there movement in the scene but the camera itself is on the 24th floor balcony in the wind!

View attachment 162923218b81403587503d5d16560e4f.jpg
Adobe Super Resolution from the 20MP ORI

There seems to be sharpening of shot noise in the image giving a strange look to some areas. Perhaps I should use Denoise first followed by Super Res.

That hasn't worked out well for me. It's as if the AI-generated artifacts generated by the denoise pass are just compounded by artifacts added by the superresolution pass. I've only tried it on a few shots, so results may vary.

Good point.

In any case, for times I can use HR or HHHR, I just assume do so. For times I can't, it's nice to know I may be able to get a little boost from the software side. They've come a long way, that's for sure.

I am curious with your FFT approach if it applies more to the older implementations of pixel shift like the EM5II and PENF that had such slow readout between frames. I rarely see the jaggies anymore once we got into the stacked sensors.

 

[Regor250]

How did you post-processed the HR/HHHR images? This makes a huge difference on the level of details one gets out of them.

I did a similar exercise a few years with Topaz Gigapixel back: https://www.dpreview.com/forums/post/66678119

And the effect of appropriate post-processing: https://www.dpreview.com/forums/post/67976275

--
Roger

 

[Asherax]

I have to say the Adobe Super resolution is quite impressive. In regards OM Systems recent pronouncement, I am in line with their assessment that Computational Photography is going to become the single largest contributor to IQ advancement.
-
I know this is a controversial statement, as like others, I seek to retain authenticity in my work, and the idea that imaginary pixels are being created to resolve issues with image quality, such as DR, IQ resolution, seems anathema to that wish.
-
That said, we already correct CA in camera, and through PP, and lens aberrations are routinely corrected in camera, without so much as a hint of dispute. I suggest that it is likely that the kinds of resolution and DR issues that are likely to come to the fore and the consequential improvements that AI and computational photographics afford will be welcomed wholeheartedly once the results become apparent.
-
The proof of the pudding is in the eating as they say.
-
The most likely situation is that computational enhancements allow smaller sensors to compete if not outstrip larger sensors in practical use, and the consequent IQ between systems becomes indistinguishable, so much so that the ergonomics of those systems becomes far more of a selling point than IQ ever was. M43's is then positioned in such circumstance to enjoy a real renaissance, given it's ergonomic and weight to FL advantage, along with it's price point for superior manufacture of smaller optics.

--
Photography is poetry made visible; it is the art of painting with light!

 

[James Stirling]

How did you post-processed the HR/HHHR images? This makes a huge difference on the level of details one gets out of them.

I did a similar exercise a few years with Topaz Gigapixel back: https://www.dpreview.com/forums/post/66678119

And the effect of appropriate post-processing: https://www.dpreview.com/forums/post/67976275

I opened the hr shot in workspace as much as I dislike the program it does a better job withe the computational stuff. Exported as a TIF for minor tweaks in PS

 

[James Stirling]

I have to say the Adobe Super resolution is quite impressive. In regards OM Systems recent pronouncement, I am in line with their assessment that Computational Photography is going to become the single largest contributor to IQ advancement.

I think so unless there is some unheard of new sensor technology , we must be very near what can be done with current tech. Which is fair enough as there is an abundance of great gear to pick from.

-
I know this is a controversial statement, as like others, I seek to retain authenticity in my work, and the idea that imaginary pixels are being created to resolve issues with image quality, such as DR, IQ resolution, seems anathema to that wish.

There is so much pre-processing going into images as you mention below that it gets hard to determine where real begins and ends. Lots of folk seem very happy with NR and it has moved on a pace in recent years but it is not a panacea. It is a lot easier to get rid of noise than to try and find any extra real detail

Assuming an appropriate scene with little or no movement the tripod pixel shift is of course giving genuine results, higher resolution, cleaner colours etc

-
That said, we already correct CA in camera, and through PP, and lens aberrations are routinely corrected in camera, without so much as a hint of dispute. I suggest that it is likely that the kinds of resolution and DR issues that are likely to come to the fore and the consequential improvements that AI and computational photographics afford will be welcomed wholeheartedly once the results become apparent.
-
The proof of the pudding is in the eating as they say.
-
The most likely situation is that computational enhancements allow smaller sensors to compete if not outstrip larger sensors in practical use, and the consequent IQ between systems becomes indistinguishable, so much so that the ergonomics of those systems becomes far more of a selling point than IQ ever was.

I agree the reason why m43 have been pioneers of these features are to compensate for disadvantages inherent in the smaller sensor size.

M43's is then positioned in such circumstance to enjoy a real renaissance, given it's ergonomic and weight to FL advantage, along with it's price point for superior manufacture of smaller optics.

Unfortunatly for m43 I think the ship has sailed , often when you can find genuinely equivalent lenses. Neither the size /weight or price favours m43. m43 is never going to be one of the big brands but it does not need to be , it just needs to provide a sustainable market that allows for the companies to continue to produce

--
Jim Stirling:
"To argue with a person who has renounced the use of reason, is like administering medicine to the dead." - Thomas Paine
Feel free to tinker with any photos I post

 

[James Stirling]

Interesting Jim, thanks for sharing. Now that Adobe puts Super Resolution (along with raw details and denoise) in LR, I find it a lot more useful (accessible?). Although the results are less persuasive than with your studio scene examples.

In my experience, the benefits of Superresolution

I am not very familiar with Canon cameras , how did their in camera upscaling perform in the R5 II is it similar in method to the Adobe . Presumably with the full software processing being done by Canon the should I imagine be able to do a better job than Adobe

over resampling using the Preserve Details 2.0 option are mixed. Where Superresolution works, it tends to do a better job with improving fine detail acutance. However, it's also more prone to generating ugly and visible patches of what look like over-sharpened/over-saturated fine detail gone awry. It's pretty random and unpredictable, which means you have to carefully examine the whole image to see if these patches show up.

View attachment 9f79b869088941aea01b841028f365cc.jpg
PEN-F pixel shift

FYI, I'm still playing with using FFTs to reduce the motion and processing artifacting in pixel shift shots. Haven't decided if the improvements are easy enough to generate on a repeatable basis to merit an in-depth posting about it. The above pixel shift shot is full of "interesting" artifacts that would make it a good candidate for analysis if you want to forward the raw to me.

View attachment 162923218b81403587503d5d16560e4f.jpg
Adobe Super Resolution from the 20MP ORI

There seems to be sharpening of shot noise in the image giving a strange look to some areas. Perhaps I should use Denoise first followed by Super Res.

That hasn't worked out well for me. It's as if the AI-generated artifacts generated by the denoise pass are just compounded by artifacts added by the superresolution pass. I've only tried it on a few shots, so results may vary.

In any case, for times I can use HR or HHHR, I just assume do so. For times I can't, it's nice to know I may be able to get a little boost from the software side. They've come a long way, that's for sure.

--
Jim Stirling:
"To argue with a person who has renounced the use of reason, is like administering medicine to the dead." - Thomas Paine
Feel free to tinker with any photos I post

 

[cathal gantly]

I spent a little bit of time at the weekend looking at Super Resolution feature in the latest version of Lightroom. I've also tried the High Res mode on my EM1.ii previously. (The EM1.ii doesn't have HHHR).

I found the Adobe Super Resolution to be very impressive, and I'm keen to spend more time investigating its capabilities, and limitations. Initial impressions are however very positive, and capable of giving the High Res modes a run for their money. I tried it on files from my EM1.ii and E-PM1 and was delighted with the results, all be it from a sample of a few images.

I think with this type of computational photography, a stand alone computer has the edge over a camera.

 

[James Stirling]

I spent a little bit of time at the weekend looking at Super Resolution feature in the latest version of Lightroom. I've also tried the High Res mode on my EM1.ii previously. (The EM1.ii doesn't have HHHR).

I found the Adobe Super Resolution to be very impressive, and I'm keen to spend more time investigating its capabilities, and limitations. Initial impressions are however very positive, and capable of giving the High Res modes a run for their money. I tried it on files from my EM1.ii and E-PM1 and was delighted with the results, all be it from a sample of a few images.

I think with this type of computational photography, a stand alone computer has the edge over a camera.

I was surprised by the results myself if you already use LR or PS it is handy and quick. It does not have all the advantages of the pixel shift modes. But you can apply it to any image even with fast movement

--
Jim Stirling:
"To argue with a person who has renounced the use of reason, is like administering medicine to the dead." - Thomas Paine
Feel free to tinker with any photos I post

 

[cathal gantly]

I spent a little bit of time at the weekend looking at Super Resolution feature in the latest version of Lightroom. I've also tried the High Res mode on my EM1.ii previously. (The EM1.ii doesn't have HHHR).

I found the Adobe Super Resolution to be very impressive, and I'm keen to spend more time investigating its capabilities, and limitations. Initial impressions are however very positive, and capable of giving the High Res modes a run for their money. I tried it on files from my EM1.ii and E-PM1 and was delighted with the results, all be it from a sample of a few images.

I think with this type of computational photography, a stand alone computer has the edge over a camera.

I was surprised by the results myself if you already use LR or PS it is handy and quick. It does not have all the advantages of the pixel shift modes. But you can apply it to any image even with fast movement

It's the post-capture ability that's particularly appealing. I've used it with files from my (long gone Nikon D100, the E-PM1 and E-M1.ii. My (so far) limited testing has produced very satisfactory results. I need to investigate it in more depth, and hope to do that soon, but on the basis of what I've seen right now, I'm questioning if I do need that S1R after all. Of course, the feature can be used on those files also...

 

[James Stirling]

I have been messing about with ACR , so going to extremes this is a 100% crop comparison of the note from the test scene . Standard raw file ( OM-1) , pixel shift and pixel shift with ACR

Test scene to show area cropped ( down sized )



Standard, Pixel shift and Pixel shift with ACR super res 100% crops



The pixel shift + ACR super resolution give a near 70" wide 300 ppi image handy for wallet photos of the kids



It is hardly perfect but you can have some fun with software these days

--
Jim Stirling:
"To argue with a person who has renounced the use of reason, is like administering medicine to the dead." - Thomas Paine
Feel free to tinker with any photos I post