[kh1234567890]

Seems to me the "laws of physics" would prevent creating more pixels out of fewer?
What am I missing?

In the realms of consensus reality laws of physics do not apply. What you get is what the AI net calculates that ought to be there if you had more pixels.

At least Adobe limits you to 2x. With Topaz there is no such limit and things can get pretty wierd ...

---

Last edited by kh1234567890; 03-16-2021 at 10:24 AM.

[DonV]

Pretty cool, makes kind of an "imaginary" picture from a "bad" picture.

[BarryE]

I'm using a newer I7 quad core processor with 128mb of ram, a 4K capable video card and a 16mp dng file took over 5 minutes to enhance in acr. Now where this does pay off is either older lower mb photos or heavily cropped photos in ACR to bring back up the resolution so I am finding in my playing with it so far.

Edit: took a K1 the same unedited DNG file and enhanced it in Lightroom Classic acr, it took 2 min to process.

The latest Lightroom Classic version created a 7360 x 4912 (Same size as original) 145.60 MB file from a 36mb size where as the latest Photoshop CC Acr created a 14720 X 9820 447.60 MB file. So looks like processing time depends on what size file the program is creating and for some reason the enhanced in Lightroom didn't increase the MP dimension. Both programs were updated this am along with ACR 13.2

I tried this with 3 different photos with the same result. Seems like it really is only working as it should currently in Photoshop ACR.

So you may want to check your picture dimensions on that rapid enhancement to see if it actually increased the resolution.

I ran it from ACR (I thought LR wasn't supported yet). The file sizes are as expected, 14720 x 9824.

[RobA_Oz]

In the realms of consensus reality laws of physics do not apply. What you get is what the AI net calculates that ought to be there if you had more pixels.

At least Adobe limits you to 2x. With Topaz there is no such limit and things can get pretty wierd ...

That little kid reminds me of Max Headroom, or perhaps Kryten from Red Dwarf! Seriously, I had been wondering about Topaz Gigapixel AI, as I’ve been using their Sharpen and Denoise AI plug-ins for a while, with generally useful results. All this means is that they haven’t enough data on profiles, even if the full-face data works pretty well. I think I’ll wait a while longer.

[gatorguy]

That little kid reminds me of Max Headroom, or perhaps Kryten from Red Dwarf! Seriously, I had been wondering about Topaz Gigapixel AI, as I’ve been using their Sharpen and Denoise AI plug-ins for a while, with generally useful results. All this means is that they haven’t enough data on profiles, even if the full-face data works pretty well. I think I’ll wait a while longer.

You don't have to accept the defaults Rob. I very often modify settings tp avoid those oddities.

[MossyRocks]

At least Adobe limits you to 2x. With Topaz there is no such limit and things can get pretty wierd ..

zoom in and enhance

[Benedict Michael]

Interpolation is just a fancy word for made up and there are tools that have been doing similar, possibly better, jobs of interpolation for years. A couple that come to mind are the freeware A Sharper Scaling, and Topaz Labs Gigapixel AI

Personally I would love to see a comparison of various upscaling methods using different programs. A good test would be to start with a pixel shift image with tons of detail. Then show a 100% crop of that and upscale it with the standard photoshop/GIMP methods (photoshop has preserve details Gimp has lanczos so both should be tested), the new super resolution feature in ACR/Lightroom, the use of the freeware tool A Sharper Scaling, the use of Topaz Labs Gigapixel AI. Scale the 100% crop up 200% with each method so we can see the results. I would do it myself but I run Linux with a windows 7 VM and don't have Lightroom, Photoshop, or Gigapixel AI so I can't do it but would love to see a good comprehensive test like that.

Been using ON1 Perfect Resize nee "Genuine Fractals" for 10 years. Long ago, I read they convert Bit Map Image to Vector, scale (up or down) and convert back to Bit Map. It works pretty good.
Can use stand alone, so not "locked in to Adobe". And while Pixel Shift is very cool, stuff that moves is a problem.

[RobA_Oz]

You don't have to accept the defaults Rob. I very often modify settings tp avoid those oddities.

Yes, I play with the variables on the other two, as well. One of the nice things about the Topaz apps is the frequency with which they provide updates, so I expect they’ll address more of the oddities that occur, over time. I still use CS5 for my basic editing, and Topaz plug-ins have boosted its capabilities quite nicely, to the point where my 11 year-old iMac still runs things quite nicely, and I don’t feel the need to upgrade to a later version (although I did buy ON1 recently on special, just to see how well it works on my new MacBook Pro).

[kh1234567890]

That little kid reminds me of Max Headroom, or perhaps Kryten from Red Dwarf! Seriously, I had been wondering about Topaz Gigapixel AI, as I’ve been using their Sharpen and Denoise AI plug-ins for a while, with generally useful results. All this means is that they haven’t enough data on profiles, even if the full-face data works pretty well. I think I’ll wait a while longer.

Faces are always going to be the hardest to make up because as humans we are pretty good at knowing when they are wrong. Less so with bushes, feathers, lawns and other random objects. Noise is also always going to be a problem.

[boriscleto]

Been using ON1 Perfect Resize nee "Genuine Fractals" for 10 years. Long ago, I read they convert Bit Map Image to Vector, scale (up or down) and convert back to Bit Map. It works pretty good.
Can use stand alone, so not "locked in to Adobe". And while Pixel Shift is very cool, stuff that moves is a problem.

Genuine Fractals nee On1 Resize has been around since 1996.

As of version 7.0, the product was called Perfect Resize, and as of version 10, ON1 Resize. There are two main features in the Genuine Fractals plug-in. First is a feature to save image files in either FIF (Fractal Image Format) or its proprietary STN multi-resolution wavelet format. This format offers file compression ratios around 2:1 for lossless and 5:1 for visually lossless. The second main feature of Genuine Fractals is a scaling algorithm based on the use of PIFS (partitioned iterated function systems). When scaling up, Genuine Fractals exploits the self-similarity of an image to increase its size while preserving detail.

Iterated function system - Wikipedia

[dadipentak]

So the idiot-proof version should be available in 3 years?

[Benedict Michael]

Genuine Fractals nee On1 Resize has been around since 1996.



Iterated function system - Wikipedia

Had no idea, didn't research the history, just my experience. I use as PS/CC plugin. Open PSD in it, scale and return to PS/CC with newly sized PSD.
Then make JPEG Print files from that. I've zero experience with the other file formats you mentioned. Works for me.

[RobA_Oz]

Faces are always going to be the hardest to make up because as humans we are pretty good at knowing when they are wrong. Less so with bushes, feathers, lawns and other random objects. Noise is also always going to be a problem.

Apparently, Tom Cruise’s face wasn’t hard to make up.

[gatorguy]

Been using ON1 Perfect Resize nee "Genuine Fractals" for 10 years. Long ago, I read they convert Bit Map Image to Vector, scale (up or down) and convert back to Bit Map. It works pretty good.
Can use stand alone, so not "locked in to Adobe". And while Pixel Shift is very cool, stuff that moves is a problem.

Gosh, I had kinda forgotten how long that software has been out. Pretty sure I was using it as far back as the 90's when I bought my first large-format printers, HP2500CP's
I can be a packrat, and sometime last year trash-canned the last very old HP found stuffed way in the back of a storage room.

[MossyRocks]

Been using ON1 Perfect Resize nee "Genuine Fractals" for 10 years. Long ago, I read they convert Bit Map Image to Vector, scale (up or down) and convert back to Bit Map. It works pretty good.
Can use stand alone, so not "locked in to Adobe". And while Pixel Shift is very cool, stuff that moves is a problem.

I knew there were other programs for upscaling but couldn't remember them. The reason I wanted to start with a pixel shift image first is that would remove the first interpolation of the demosaicing process. Since I would be going for a test shooting off of a tripod with an exceptionally sharp lens with good light and stopped down to the ideal point at base ISO. This would give a good starting point for the best comparison of the different tools.

Maybe I will spend some time later tonight and shoot a pixel shift image with the DFA 100/2.8 WR macro at f/5.6 of some setup with lost of fine detail and start the upscaling test with what I have available to me and post a link to the dng file (google drive) so others can use the same test image and use their upscaling tool to produce similar results.