An upsampling algorithm that works - ML Super Resolution in Pixelmator Pro

This is the best upsampling algorithm I've seen yet. It doesn't add detail, because information theory says that you can't recover information that has been destroyed. What it does, however, is dramatically clean up low resolution images to make them able to be enlarged.

 

Link: All about the new ML Super Resolution feature in Pixelmator Pro - Pixelmator Blog

 

It's not a competitor to a high resolution image. But at worst, it's better than nothing. Imagine if you have an important image, and you lost the high res version of it. You can now make some practical use of the low res image.

It's interesting to note that the rain-on-window examples don't use the same shot/framing. Hmmmm... The example of the portrait doesn't show what this program would do to the eyes, the most important feature of a portrait. Hmmmm...

Does it run on Macs only or also on Windows machines?

The images use the common before/after viewer. You have to drag the line down the middle back and forth to see the comparison. Drag the line to the left of the eyes and you see the effect on the eyes. (I don’t love the effect on the eyes.)

 

Yes, Pixelmator is Mac only.

Impressive. Unfortunately, I don't use a Mac, so I can't test it for myself.

 

It doesn't add detail, because information theory says that you can't recover information that has been destroyed.

 

That's actually not right. All of the methods they show add information. That's the core of how they work. They don't add back original information; that is lost and can't be recovered. Instead, they use various algorithms to impute additional data points--that is, to make up additional detail based on mathematical computations.

That's actually not right. All of the methods they show add information. That's the core of how they work. They don't add back original information; that is lost and can't be recovered. Instead, they use various algorithms to impute additional data points--that is, to make up additional detail based on mathematical computations.

I do hate to nitpick but my statement was correct: once the original information has been destroyed, you can't recover it. ;-)

High resolution image without more detail????

I have to give it to you Karim.

You're both correct. There is more information, in the sense that there are more pixels. There is not more information, in the sense that the upsized image does not convey any more details of the actual scene than the original.

What you originally wrote was this:

 

It doesn't add detail

 

not this:

 

once the original information has been destroyed, you can't recover it

 

The original statement is wrong, while the second is correct, as I explained in post #4.

 

To get more resolution, one has to add detail. There is no other way. Because the original data are lost, the software has to generate imputed values. The issue is how well it does that.

High resolution image without more detail????

I have to give it to you Karim.

Do pay attention, BeBu. ;-)

 

The original statement is wrong, while the second is correct, as I explained in post #4.

I see. Fair point!

Do pay attention, BeBu. ;-)

 

OK I must give you an award.

you can't recover information that has been destroyed

I don't know who (or what book) said that but they can recover past information that has been destroyed, they also can discover new information that nobody ever knows and they even can predict information about what will be in the future (not happens yet). Of course, they don't know everything and they are not always right. Need examples? : they know what is the temperature tomorrow where you live, whether you will have a white Christmas or not, was a 10 years old boy was conceived by Mr. X and Ms Y? and the examples can go on forever. You can delete your files and reformat your hard drive, but they can "recover" a lot of information that was in it. The only thing is how to improve the algorithm to be applied for more kinds of information and higher accuracy. A "lossy" compressing algorithm can create a much smaller data file; i.e throw away a lot of information, but it still can recover most of it. You have seen a lot by using JPG's and hear a lot by using MP3's and not many people can realize the difference.

I don't know who (or what book) said that

You and I are talking about very different things.

 

By definition, information that is destroyed cannot be recovered. This is a logical statement that you should find in any text book on information theory.

 

However, information that is thought to be destroyed can actually still exist. They found a cut of Blade Runner that was thought to have been destroyed, just to give one example. And thanks to someone's diligence (and possible anal retentiveness!), there exists a cut of Teen Wolf which apparently only exists on a home recorded VHS tape. Things like this happen often. Maybe there is hope for the Nazi gold and art treasures that are still missing.

 

Predicting the future is very difficult and unreliable, contrary to what you may have been told. That is, unless you apply a good intuition or even psychic ability, but those things are a separate discussion entirely.

 

Weather and climate models are accurate for short terms only. You do not know what the temperature will be this time next year. But you can take a good guess, of course. The CMIP is hopeless, due to poor assumptions. Maybe it can work with better input? I am open to that.

 

Have you ever tried to compress a text file as a JPEG? I don't mean converting it into a bitmap - I mean, take the ASCII text and compress it with the JPEG algorithm. Tell me what you get.

to compress a text file as a JPEG?

nobody said that the same algorithm can be used for anything.

Weather and climate models are accurate for short terms only

every model/algorithm has a limit (usually very restricted) like in the example they increase the resolution by 200%. But then you admit that they can predict your temperature tomorrow. Of course my wife (or even myself) can not predict whether I would buy another camera for this Christmas.

what you may have been told.

I did not say what I have been told, I was just trying to tell you something (but maybe I should stop if you would not listen)

By definition, information that is destroyed cannot be recovered

matter can neither be created nor destroyed, though it can be changed from one form to another). The only hard thing (that the ingenious algorithms do) is to return to the original form.

matter can neither be created nor destroyed, though it can be changed from one form to another). The only hard thing (that the ingenious algorithms do) is to return to the original form.

With respect, we are not talking about matter. We are talking about information.