Jump to content

Topaz Sharpen AI


Recommended Posts

OK, but just for fun I sharpened the right pane with the free program GIMP...not so sure what is so special about Topaz. please enlighten me.

Two things I notice about the GIMP sharpening:

 

The righthand frame of the window itself, not the outer frame but the inner frame, is still blurry and now looks completely off from the rest of the pic. The lefthand inner frame of the window itself is also still a bit blurry. These may be more a matter of execution than software.

 

The sharpening of the curtains themselves is more extreme and looks more software-generated and artificial than the TOPAZ sharpening. It results in odd patterning and no longer has the softness of lace behind glass. Again, this could be more a matter of execution than GIMP.

 

Truth is, with the TOPAZ sharpening, which looks better than the GIMP, there's still softness on the inner righthand frame of the window itself. When looking carefully, both seem to have been artificially sharpened.

 

Certainly worth a shot and perhaps more careful execution would be more organic, but probably best to focus the camera ... if possible.

  • Like 3

"You talkin' to me?"

Link to comment
Share on other sites

Two things I notice about the GIMP sharpening:

 

The righthand frame of the window itself, not the outer frame but the inner frame, is still blurry and now looks completely off from the rest of the pic. The lefthand inner frame of the window itself is also still a bit blurry. These may be more a matter of execution than software.

 

The sharpening of the curtains themselves is more extreme and looks more software-generated and artificial than the TOPAZ sharpening. It results in odd patterning and no longer has the softness of lace behind glass. Again, this could be more a matter of execution than GIMP.

 

Truth is, with the TOPAZ sharpening, which looks better than the GIMP, there's still softness on the inner righthand frame of the window itself. When looking carefully, both seem to have been artificially sharpened.

 

Certainly worth a shot and perhaps more careful execution would be more organic, but probably best to focus the camera ... if possible.

 

The GIMP sharpening looks very overdone and crunchy.

  • Like 2
Link to comment
Share on other sites

Two things I notice about the GIMP sharpening:

 

The righthand frame of the window itself, not the outer frame but the inner frame, is still blurry and now looks completely off from the rest of the pic. The lefthand inner frame of the window itself is also still a bit blurry

That's obviously the edge of a selection marquee.

The GIMP sharpening looks very overdone and crunchy.

Not a fair comparison without access to the original. There are obviously oversharpened JPEG artefacts from the small posted sample.

 

FWIW. Spherical aberration, which is what the 55mm f/1.2 Nikkor has in spades, responds well to conventional sharpening techniques. As does a small degree of diffraction blurring.

Link to comment
Share on other sites

Two things I notice about the GIMP sharpening:

 

The righthand frame of the window itself, not the outer frame but the inner frame, is still blurry and now looks completely off from the rest of the pic. The lefthand inner frame of the window itself is also still a bit blurry. These may be more a matter of execution than software.

 

The sharpening of the curtains themselves is more extreme and looks more software-generated and artificial than the TOPAZ sharpening. It results in odd patterning and no longer has the softness of lace behind glass. Again, this could be more a matter of execution than GIMP.

 

Truth is, with the TOPAZ sharpening, which looks better than the GIMP, there's still softness on the inner righthand frame of the window itself. When looking carefully, both seem to have been artificially sharpened.

 

Certainly worth a shot and perhaps more careful execution would be more organic, but probably best to focus the camera ... if possible.

I didn't notice at first, until the side sharpened with the Gimp software started making me dizzy.

  • Like 1
Link to comment
Share on other sites

I did a comparison with Topaz Sharpen AI and photoshop and honestly I couldn't tell the difference except that PS may have taken a little more tweaking to produce equivalent results.

One significant drawback was that, unlike PS, TopazAI was profoundly slow, to the point of being virtually useless - I could almost make a cup of coffee and drink it while waiting for it to process a Nikon D850 image. The machine I use is an iMac with 3.4GHz quad-core 7th-generation Intel Core i5 processor, Turbo Boost up to 3.8GHz, 32GB RAM. So make sure you read the TopazAI hardware recommendations and "try before you buy".

  • Like 1
Test
Link to comment
Share on other sites

I've been playing around with the app some today. I have it set-up as a plug-in for Photoshop, and I would say having it come in as a layer that I can selectively brush the effect out is a necessity. I see quite a few places where the sharpening is unwelcome or simply weird. It's really slow, and it doesn't always work miracles; and compared to Photoshop's Smart Sharpening, which isn't nearly as slow, I often end up preferring the output of Smart Sharpening. I have used both Photoshop's Smart Sharpening and Topaz Sharpening in part because Topaz doesn't create as much noise in skies as Photoshop's Smart Sharpening.

 

Anyway, I have 30 days to evaluate it. I'm not sold so far because of the cost of the app, its sluggishness, and its utility. Heck, I can shoot at f/16 and when I resize for the web I have to be careful not to over-sharpen the file, so this is really just for cropping a lot or printing really big (and then those artifacts from Topaz Sharpening start to become ugly).

  • Like 1
Link to comment
Share on other sites

I can only comment on what I see, but I don't think that there is anything 'AI' going on here. It's just a sharpening algorithm that takes up a lot of CPU resources. I don't like the result. AI is not a magic shortcut around information theory. I'm not knocking the OP, just being honest, as I think we all ought to be.
Link to comment
Share on other sites

Doesn't AI mean Artificial Intelligence? How can there be AI involved in a sharpening algorithm?

 

Not that I at all know what they are doing, but a big problem with sharpening is to know how much.

 

Sharpening increases the higher spatial frequencies. Years ago, someone asked me about a sharpness

knob on a VHS player. (Yes, that long ago.) Why wouldn't someone want the most sharpness?

Why not turn the tone control all the way up on a stereo amplifier?

Because you can have too much high frequency components, which doesn't look right.

 

So I actually don't know at all what they do, but the problem is to know how much high

frequency signal to increase.

 

Well, there is a whole subject of non-linear deconvolution, in addition to the more

usual linear deconvolution.

 

My favorite book on such is Jansson's "Deconvolution of Images and Spectra".

 

For linear deconvolution, you take the Fourier transform, divide by the transform

of the believed point spread function, and then inverse transform. The problem is

that most sources have noise, and that the PSF transform might have small values.

 

One problem is that intensity should never be negative, but the math doesn't

know that.

-- glen

Link to comment
Share on other sites

I did a 'quick' (not really) test and am decidedly not impressed with Topaz Sharpen AI.

I used my 55mm f/1.2 Nikkor - same as the OP.

Here's the subject matter.

DSC_3261-small.thumb.jpg.ccf8638069300985452cf1e39933a3ae.jpg

Not something I'd usually choose that lens for, but plenty of detail.

 

The subject distance was about 1.5 metres, and not as close as the above crop looks.

 

Here are 100% crops of what the lens delivered at f/1.4.

55mm_f1-4.thumb.jpg.7f45885137b584ad4e094b812a5bb7d1.jpg

The detail all smudged by an SA halo.

Here's Topaz Sharpen's sharpening with the slider right across at 100.

479931879_55mm_f1-4-SharpenAI-sharpen(1).thumb.jpg.8b769e39765474b12044ab5639304f92.jpg

I don't see an amazing difference. So I swapped to the 'Focus' setting at 100.

55mm_f1-4-SharpenAI-focus.thumb.jpg.b2472d258fd569742a043b73485410ce.jpg

Crisper, but the lettering has been thinned, which I'm not liking, and there are some edgy artefacts on the chromework of the camera.

 

Let's try PhotoShop's Smart sharpen filter.

55mm_f1-4_PS_Smart-sharpen.thumb.jpg.a3cf2a6abeb5905c6e2975d1dfee68fb.jpg

Much better, IMO. And substantially quicker.

 

How about straightforward USM in GIMP?

55mm_f1-4_Gimp_USM.jpg.92a690867282dbf7c8f3448c265aa898.jpg

Not as good as PS Smart sharpen, but better than either of Topaz's renderings, in my view.

 

And none of them are as good as just stopping the lens down to f/2.8!

55mm_f2-8.thumb.jpg.25f71e0b910ee0c62398e8cac4112efa.jpg

OK it's not lace curtains, and maybe there are subjects where Topaz Sharpen AI will do better.

Make of it what you will, but that software isn't on my shopping list I'm afraid.

  • Like 3
Link to comment
Share on other sites

Not that I at all know what they are doing, but a big problem with sharpening is to know how much.

 

Sharpening increases the higher spatial frequencies. Years ago, someone asked me about a sharpness

knob on a VHS player. (Yes, that long ago.) Why wouldn't someone want the most sharpness?

Why not turn the tone control all the way up on a stereo amplifier?

Because you can have too much high frequency components, which doesn't look right.

 

So I actually don't know at all what they do, but the problem is to know how much high

frequency signal to increase.

 

Well, there is a whole subject of non-linear deconvolution, in addition to the more

usual linear deconvolution.

 

My favorite book on such is Jansson's "Deconvolution of Images and Spectra".

 

For linear deconvolution, you take the Fourier transform, divide by the transform

of the believed point spread function, and then inverse transform. The problem is

that most sources have noise, and that the PSF transform might have small values.

 

One problem is that intensity should never be negative, but the math doesn't

know that.

Thanks Glen,

I am objecting to the misuse of the term "AI", but this is not the only company guilty of that.

Unless I am understanding the product wrongly, it is basically a static algorithm using predefined criteria to sharpen, much like Nikon's MatrixMetering that uses a database of sample photographic situations to apply best possible meter matching.

It is also possible I expect too much of artificial intelligence.

Niels
Link to comment
Share on other sites

I am objecting to the misuse of the term "AI", but this is not the only company guilty of that.

Unless I am understanding the product wrongly, it is basically a static algorithm using predefined criteria to sharpen, much like Nikon's MatrixMetering that uses a database of sample photographic situations to apply best possible meter matching.

It is also possible I expect too much of artificial intelligence.

 

I don't know at all what they are doing, but I have been to seminars from people using AI methods for visual problems.

 

I am almost remembering some from that seminar. There was one where it would transform

images of houses. It had the funny result that it would put a door where there shouldn't be

a door, but the training data had doors.

 

If there is training data with lace curtains, you can make an AI algorithm that will sharpen

lace curtains. It will also make things that aren't curtains look like curtains.

 

Since there is a relatively small set of types of scenes that a large number of people photograph,

it is possible to train on those. Houses, people, and trees are common in many photographs,

so one could design one to sharpen up those, based on other houses, people, and trees.

 

So, I don't know that is what they are doing, but I do know that people are doing it.

  • Like 1

-- glen

Link to comment
Share on other sites

I am objecting to the misuse of the term "AI", but this is not the only company guilty of that.

Unless I am understanding the product wrongly, it is basically a static algorithm using predefined criteria to sharpen,

 

That's my understanding. The vendors use machine learning on large datasets to develop canned algorithms that are--ideally--more effective than those humans write. I don' think there is any AI at all going on in your local computer.

 

Sometimes it works. The AI-based selection tools in Photoshop are quite useful and in some cases really remarkable. In other cases, they don't work. My favorite example was subject selection of a woman holding a toddler. I'm guessing that they didn't have many photos like that in the training set. The algorithm immediately drew a nice selection of both people as a single subject.

 

I played a bit with the Topaz tool some time ago, using it on an image that had substantial problems of sharpness. I thought it added a tiny bit, but I wasn't impressed enough to buy it.

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...