Jump to content

Topaz Lab's AI Gigapixel Image Resize Program


Recommended Posts

Is upsizing still a thing since 24 megapixel and upwards cameras became commonplace?

 

I remember being very unimpressed with the results from 'Perfect Resize', Fractal Whatever and such offerings years ago. I don't imagine the algorithms have changed much, although powerful GPUs may have made the processing a bit quicker. There's a limit to how good you can make imaginary or guesstimated interpolated pixels look.

 

Topaz seems to work quite well on the geometric stonework, but the dog looks very faceted and CGI-ish, and the grass blades are obviously a stock texture 'pasted' in. Quite clever, but why would you? When any decent camera gives more detail than most people ever need.

Edited by rodeo_joe|1
Link to comment
Share on other sites

but the dog looks very faceted and CGI-ish, and the grass blades are obviously a stock texture

Have you seen how small and not-so-sharp the dog is in the original photo? Obviously this example was used for comparison and testing the software.

 

Btw, the current version of Perfect Resize works very well. Photoshop has come a long way as well. It just seems that Gigapixel has a small edge over them.

 

Upsizing software has its place in a photographer's toolset as many photographers still have old photos shot with lesser cameras in the past; it is also immensely useful for cropped images.

Link to comment
Share on other sites

Yes, the dog looks sharper, but it also looks less 'real' and more cartoon-like.

 

The re-constructed grass worries me. It's obviously an AI interpretation of what generic grass looks like - in the same way that a painter working from memory would fill in such an area. The adjacent rocks/stones don't get such a good enhancement, and the scene becomes discontinuous with the 'focus' seeming to shift within a space of inches.

 

So why does this worry me? I have to ask if such an AI artist's impression would stand up as evidence in a court of law. If the grass has been re-imagined, then what else is false? Is it really 100% a photographic image any longer? Or just a CGI re-interpretation of the original scene?

 

What would such a neural-net's imagination make of the Zapruda footage, for example?

 

Personally, I see nothing wrong with the interpolated upscaling samples - noise and all. OK, they're not very sharp, but at least they don't present us with the dilemma of deciding whether they contain false information or not. Nor of looking CGI rendered.

Edited by rodeo_joe|1
Link to comment
Share on other sites

I have to ask if such an AI artist's impression would stand up as evidence in a court of law. If the grass has been re-imagined, then what else is false? Is it really 100% a photographic image any longer? Or just a CGI re-interpretation of the original scene?

You worry too much. If an image is artificially altered, then the court does not use it, that's all. :D

 

Now that you mentioned it, I am more worried from a photographer's viewpoint because some of us photographers don't want to alter anything.

Link to comment
Share on other sites

You worry too much. If an image is artificially altered, then the court does not use it, that's all. :D

 

Now that you mentioned it, I am more worried from a photographer's viewpoint because some of us photographers don't want to alter anything.

 

- Exactly my point. I don't care about the legal implications, just the verisimilitude of the image.

 

As to where this sort of technology is heading; I'll let future generations worry about that.

Link to comment
Share on other sites

If you don’t want to alter anything, don’t alter anything.

It depends on what the goal is at that particular situation. There may be a photo that must be enlarged to make a decent print. Other than enlarging, in most occasions, I suppose "improvement" by the software is welcome, as long as they are not blatant alterations, which I think would not be.

Link to comment
Share on other sites

  • 2 weeks later...

This type of software is used mostly with high end or large format printing. It's not used for smatphone shooters to make their pics look like LF capture backs.

 

One advantage the RA-4 based labs have over smaller ink-jet based printers is the front end RIPs are really, really good on Lambda's, LightJet's, Fuji's etc. I've printed 40x60 murals off these things well below 150dpi, and they are tack sharp provided I'm handing them a detail full file. When you have a $10,000 software package upsizing it's makes all the difference.

 

However, with a large format ink-jet printer you are pretty much stuck with native DPI, or tricks in Photoshop which aren't that good. Yet another reason RA-4 dominates large format printing while the poor ink-jet guys have to pester clients to upsize their files ahead of time. This type of software allows you hit a 360dpi from a much lower rez file and gain the most out of the mechanical resolution of the printer. When I'm selling a large print to a client I don't want to see upsizing that's anywhere like bicubic.

 

As for the Zapruda footage that's more likely to go into some type of photoaccute type algorithm which extrapolates information from sequential images. That type of processing is used all the time in court proceedings.

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...