Gigapixel AI resizer vs FastStone.

[tbcass]

I downloaded the free 30 day trial just to see what it can do. I used a 4mp photo from my Olympus c4000, used Gigapixel to resize 2x and FastStone using Lancos 3 with some sharpening and Lancos 2 which sharpens by default. Which do you prefer? Answer to follow.















--
Tom

 

[Guy Parsons]

Needs some pixel peeping comparo but my eyeballs see #1 as "sharpest", #3 next and #2 worst, but it is all so close at realistic viewing sizes that any of those would do OK.

I use "sharpest" in quotes as #1 looks a bit weird at 100% so whatever made #1 would not be my choice, #3 for me for best compromise between sharpness and artifacts.

The bottom left corner at 100% using FastStone Viewer is what I was comparing to see differences.

 

[sybersitizen]

#1 is clearly the most detailed of the three. If that's not the Gigapixel AI version, you must have done something wrong because any kind of Lanczos interpolation should never beat it for sharpness.

 

[Karl Persson]

1=Lanczos 2

2= Gigapixel

3=Lanczos 3

And I´d prefer nr 3

 

[tex]

Disclosure: I purchased G AI for a job and have used it successfully in a major museum setting with Pentax 645Z files with splendid results.

G AI is a tricky program---it can spit out really bad results. It can also work beautifully. My advice: don't use Auto and don't try to enlarge too much. My advice to the Topaz design team: give users more controls; to the Topaz marketing team: ratchet back your promises.

Of these 3 examples, 1 and 3 have unpleasant artifacts, 2 is too soft. You have to start with a better file in the first place, I'd say....

 

[Tuloom]

IDK, but #2 is especially poor.

I also think the original file is jacked.

 

[Glen Barrington]

IDK, but #2 is especially poor.

I also think the original file is jacked.

I'm not sure what jacked means.

 

[tbcass]

For this test I picked a photo that had fine pollen grains.

First the original 3.9mp photo.





Now 3 samples resized to 15.67 MP.













--
Tom

 

[tbcass]

IDK, but #2 is especially poor.

I also think the original file is jacked.

I'm not sure what jacked means.

Me either but I just posted a second sample that starts with a better original.

 

[Glen Barrington]

Could it be that a 2X enlargement is not enough of a test? I don't think Gigapixel AI is likely to be purchased or used for casual enlargements, but more for those challenging photos and situations. In my mind, it is just too expensive for casual work.

Most of my resizing needs are of the reduction kind and not the enlarging kind, so it seems unlikely I would buy Gigapixel AI for that. Besides, I think ACDSee does an excellent job of casual resizing work.

The two most important Topaz tools in my mind are Denoise AI and Sharpen AI as they are clearly better than most built-in tools. I may eventually buy the Jpg to Raw AI as I recently saw some impressive demos. But I don't know, I rarely shoot jpg any more.

--
I'm no genius, but I play one on the internet.
http://glenbarrington.blogspot.com/
http://glenbarringtonphotos.blogspot.com/
https://www.flickr.com/photos/130525321@N05/

 

[Tuloom]

The first example is a great example of garbage in, garbage out.

Second test is much better.

#1 is the clear winner. That dog'll hunt.

 

[Glen Barrington]

At the default size, photo 2 seems marginally better, but at a 100% view, it is clearly better. My comments about the casual use of Gigapixel AI still hold I think. It's not going to be purchased for occasional mundane resizing needs.

 

[sybersitizen]

Same observation as the first test (skipping the original here as #0):

#1 is clearly the most detailed of the three. If that's not the Gigapixel AI version, you must have done something wrong because any kind of Lancos interpolation should never beat it for sharpness.

 

[tbcass]

Same observation as the first test (skipping the original here as #0):

#1 is clearly the most detailed of the three. If that's not the Gigapixel AI version, you must have done something wrong because any kind of Lancos interpolation should never beat it for sharpness.

You are absolutely correct that #1 is the Gigapixel version in both cases. #2 is Lancos 2 and #3 is Lancos 3 with some sharpening. I have come to some conclusions. The first is the original photo must be of sufficient quality because GIGO. You can't make a good photo out of a bad one. My second is Gigapixel is much better than Lancos and the excellent quality makes it a good choice for enlarging a low resolution photo for large prints. In fact the quality is much better than I expected after the poor results from my first try. All in all the AI seems much better than Sony's CIZ or what I've seen from any smartphone. The down side is it's quite slow despite the fact I am using a fairly powerful computer. I didn't time it but it was in the neighborhood of 30 +/- seconds. Another problem is the interface is quite primitive, probably because I'm using it as a stand alone.

 

[tbcass]

At the default size, photo 2 seems marginally better, but at a 100% view, it is clearly better. My comments about the casual use of Gigapixel AI still hold I think. It's not going to be purchased for occasional mundane resizing needs.

Are you sure about that? To my eyes the Gigapixel version is much better than #2 or #3. It's not even close.

 

[Glen Barrington]

At the default size, photo 2 seems marginally better, but at a 100% view, it is clearly better. My comments about the casual use of Gigapixel AI still hold I think. It's not going to be purchased for occasional mundane resizing needs.

Are you sure about that? To my eyes the Gigapixel version is much better than #2 or #3. It's not even close.

Pretty sure, I took some time comparing and re-comparing. I like #2 better. Again, I don't think a 2X enlargement is really testing Gigapixel AI. for all we know, a simple 2X enlargement in GP Ai just might use Lanczos for a non-difficult photo, and FS has tweaked Lanczos a bit better. Certainly, I don't think there is enough difference to warrant a clear win to GP Ai. I think a 4X test might be more of a challenge, maybe I'll do one of my own.

It's not like any of us don't have the time!

--
I'm no genius, but I play one on the internet.
http://glenbarrington.blogspot.com/
http://glenbarringtonphotos.blogspot.com/
https://www.flickr.com/photos/130525321@N05/

 

[tbcass]

Check out these 2 crops. To me the Gapixel version looks much better at 100%.


Gigapixel Crop


Lancos 2 crop

--
Tom

 

[tex]

That's a laugh. Try 40 minutes on my rig. Of course, I was starting with 645Z tiffs....

I would also add that G AI is not just good for milking low mp shots, but for getting extremely large prints from high mp files---the ones I did were at least 60" on the long side, and the situation for display (in a vitrine) was such that people would be looking very closely at the prints with some of the actual items depicted laying on top. Rez was perfect.