Mike Chaney's Tech Corner
November 29, 2024, 04:49:19 PM *
Welcome, Guest. Please login or register.
Did you miss your activation email?

Login with username, password and session length
News: Qimage registration expired? New lifetime licenses are only $59.99!
 
   Home   Help Login Register  
Pages: [1]
  Print  
Author Topic: In Depth Look at Topaz's new A.I. Gigapixel Interpolation (20 minute video)  (Read 7493 times)
admin
Administrator
Forum Superhero
*****
Posts: 4224



Email
« on: September 05, 2018, 04:39:26 PM »

Lots of people are talking about the new A.I. Gigapixel interpolation tool and reviews are popping up everywhere.  So naturally, I'm getting questions from Qimage users.  Questions like "Have you reviewed it?"  "How does it compare to Qimage's interpolation or other math based resampling?"  "If it's that good, should I just use it on ALL my photos prior to printing with Qimage?"  Or even, "I'm trying it but it's taking 30 minutes to resample one photo".

In this video, I do a quick and simple explanation of interpolation and go into what the A.I. interpolation is doing behind the scenes.  I look at some pros and cons of mathematical versus A.I. methods and do a lot of pixel-peeping to review results of the A.I. Gigapixel tool versus Qimage's fusion interpolation.

So if you want to geek out and do some pixel-peeping to see what this new interpolation tool has to offer, feel free to check out the video:

https://youtu.be/LGWEyG4DUOM

Regards,
Mike
Logged
Alain
Newbie
*
Posts: 7


« Reply #1 on: September 05, 2018, 09:43:43 PM »

Lots of people are talking about the new A.I. Gigapixel interpolation tool and reviews are popping up everywhere.  So naturally, I'm getting questions from Qimage users.  Questions like "Have you reviewed it?"  "How does it compare to Qimage's interpolation or other math based resampling?"  "If it's that good, should I just use it on ALL my photos prior to printing with Qimage?"  Or even, "I'm trying it but it's taking 30 minutes to resample one photo".

In this video, I do a quick and simple explanation of interpolation and go into what the A.I. interpolation is doing behind the scenes.  I look at some pros and cons of mathematical versus A.I. methods and do a lot of pixel-peeping to review results of the A.I. Gigapixel tool versus Qimage's fusion interpolation.

So if you want to geek out and do some pixel-peeping to see what this new interpolation tool has to offer, feel free to check out the video:

https://youtu.be/LGWEyG4DUOM

Regards,
Mike

Thanks

I understand that it needs visual inspection afterwards and thus not for a "automatic" running printjob.
Logged
Terry-M
The Honourable Metric Mann
Forum Superhero
*****
Posts: 3251



WWW
« Reply #2 on: September 07, 2018, 04:29:15 PM »

Quote
So if you want to geek out and do some pixel-peeping to see what this new interpolation tool has to offer, feel free to check out the video:
I did and found your video very interesting , thank you!
Maybe eventually the AI method will have a better data base to work from but the slow speed seems to be a big disadvantage.
Terry
Logged
admin
Administrator
Forum Superhero
*****
Posts: 4224



Email
« Reply #3 on: September 07, 2018, 05:20:59 PM »

Quote
So if you want to geek out and do some pixel-peeping to see what this new interpolation tool has to offer, feel free to check out the video:
I did and found your video very interesting , thank you!
Maybe eventually the AI method will have a better data base to work from but the slow speed seems to be a big disadvantage.
Terry

Yeah, maybe it'll improve over time.  I think it's a good option when you really need to push the pixels, even if those times may be rare these days with 24+ MP cameras.  I honestly can't think of a time in the last few years where I've had a crop tight enough or a print big enough where I'd need something that heavy handed, but when trying to rescue an older or cell phone photo, it's a good option.  It works well on "random" scenes like sunsets, landscapes, flowers, etc. but things like line art, faces, geometric patterns like wheels, can show "weirdness" because you are familiar with what they should look like.  Meaning, you won't notice a random tree branch out of place, but the wrong eye color or mouth shape: that stands out.

Speaking of improving over time, I'm not clear on how they are doing their A.I. and whether or not they have an online database or a local one that just improves with each version.  Either way, that brings up the issue of consistency and reproducibility: if you use it on the same image a week from now, will you get the same results?  Not sure about that.  I tried the new 1.1 version and while the same areas had similar anomalies, the anomalies looked "randomly different".

My suspicion is that you'll always be able to pick out some anomalies in recognizeable subjects because in order to keep the results sharp, they have to pick what type of missing information to insert... and there's just not enough data to determine that precisely.  So the end result is that you either end up with something that is a bit soft (as with the math methods) or sharp but with some artifacts (A.I.).  I do think the tool mitigates those artifacts quite well, so it's not like they ruin the image.  Let's face it, by the time you get to the point where you see artifacts (or softness), you're really pushing the limits.  And if you are NOT pushing the limits... you probably don't really need heavy handed methods anyway.  Kind of a catch 22.

Mike
« Last Edit: September 07, 2018, 05:22:44 PM by admin » Logged
Jeff
Hero Member
*****
Posts: 764



WWW Email
« Reply #4 on: September 08, 2018, 07:44:57 AM »

I doubt I will ever need this level of up sampling, but a very good understandable video explaining the process.

Thanks, all adds to our general knowledge.

Jeff   
Logged

Grumpy
Pages: [1]
  Print  
 
Jump to:  

Powered by MySQL Powered by PHP Powered by SMF 1.1.21 | SMF © 2015, Simple Machines Valid XHTML 1.0! Valid CSS!
Security updates 2022 by ddisoftware, Inc.