Topaz Photo AI v3.0.3

Shouldn’t be doing that

1 Like

Agreed, that’s why I posted it in this thread. Since I started using PAI version 1.3.1 I have never had this occur until today.

@dakota.wixom

It simply does not work that way.

When using the Pixelshift, you don’t get more resolution in detail … not that much.

Because the pixels don’t change its size by 2x, but its position.

So you get more resolution in color, tones and less Noise.

And the images does get more flat a little bit (at 1:1) because of the “more” signal you get.

And i think the diffraction does step in at a very early point.

The pixelsize of the GFX 100 II is at 0,0037 mm, that means that at a aperture of f4.0, 4 sourrounding pixels are already reading the detail of a single pixel in the center of them, and because of this the resolution in detail does already shrink.

With deep learning, the AI naturally learns the details that are not available (because of diffraction) and because AI is an equation, the effect is likely to be even greater.

I then had the idea that there should be an AI that determines the position of a pixel and another AI that determines the brightness.

Because in reality each pixel is exposed independently of the others, but with AI each pixel is always dependent on the others.

My goal would be to create high-resolution images that also look “natural” (in the sense of what we photographers are used to) and are not smeared with equations.

I think the smear of independent details is what bothers me most about AI.



This is how you get a raw image presented when you load it into an (any) raw converter (because if not changed they load its base profile), its already denoised and lens corrected (diffraction)



This is the output i would like to have to work with after PhotoAI enlarge or Gigapixel.
An at pixellevel independent output.
Thats at 300%

You do if your lens can resolve it.

I think the 120mm Macro GF lens can do it (almost 3K lens).

And his images show the same effect.

You have better tones and colors, but almost no gain in detail resolution, because pixelshift does not place you closer to your subject, does not change the pixel size and becaouse of this does not make the sensor denser.

The increase in resolution does come from the additional exposure of the pixels.

I do not think so based on those materials, it does make your sensor dense. E.g. from the second link:

"The single-shot DNG version has a peak resolution of 77.6 lp/mm at 5504 pixels tall. The DNG version of the pixel-shifted 16 merged shots has a resolution of 74 lp/mm at 11008 pixels tall. So why in the world does the pixel-shifted shot seem to have slightly lower resolution? Because it has twice as many equivalent pixels in both the horizontal and vertical directions!

Another way to express resolution is in units of line pairs per picture height (lp/ph), where you multiply the line pairs per millimeter by how many millimeters tall the sensor is. With pixel-shifting, you essentially double the number of millimeters in the sensor, so the Z8 sensor would change from 23.9X35.9 to 47.8X71.8 millimeters! This means the resolution changed from 1855 lp/ph to 3537 lp/ph! Definitely improved resolution! The percent change in resolution is actually about 90 percent! This resolution is the equivalent of an MTF50 148 lp/mm from a non-pixel-shifted sensor with the 23.9X35.9 dimensions."

And the following Real Life Example in the same article.

Yes i know, i did read it too, but thats not what i’ve seen on the images i’ve taken.

I’ve read in a forum that just use Gigapixel for this does result in almost the same image as with pixelshift.

Too bad, at your link, the user does present only imagea of a software and not something you can look at for yourself.

Yes, there is the image with the Bottle, but the output you get has only finer lines but you cant see more detail.

More detail means to me:

Clear seperated from each other detail, like hairs or sand.

The thing i did photograph was fibres from a backpack and they only did grow in size but the fibres did not tear from each other but did stay the same.

I know that you photographed that backpack :slight_smile: Am not sure if the settings for pixel shifting in your camera were optimum for increased resolution. E.g., in the Nikon Z8 tutorial that I posted above only the options with 16 and 32 repeat images give the real increase of the pixel “density” in the resulting image, whereas other just improve noise and colors, as you said.

Anyway, I do not see a situation where I would need or want to use this feature all things considered.

On another note, I just recently tried focus stacking in Photoshop, from just manual macro shooting, and was really impressed by the result. No relatedness to Photo AI or increased resolution of the sensor but the amount of sharpness, fine detail, and denoising is incredible. Yet, makes sense only for macro as I see this feature, but this is something that I am going to use from now on…

It appears there is still an issue with Photo AI trying to write verbose logs (or whatever it is doing) after processing larger batches. It becomes unresponsive to the point of requiring a force quit on Mac M2.

Is anyone else seeing this? I remember @david.123 and several others called this out many versions ago!

I’m sure it’s still a problem as I have not seen an option to disable logging added as per my request:

Thanks for the reminder to start pestering them about this!

@dakota.wixom is this request at least on the planned features list?

1 Like

The long-standing issue with TPAI being unable to save edited images back to Apple Photos in certain common scenarios is still there. Will it ever be addressed?

See e.g. here:

It was set to 16 images, 16bit RAW.

I just wanted to share what i’ve seen and how it could influence deep learning/ machine learning applications in case someone wants to train his models with the output.

Understood and much appreciated!

Hi i have been a topaz user for a while now but the latest update to Topaz AI 3 is proving to be a nightmare and can not get help from Topaz they told me to upgrade my Graphics card which is the second time i have done in 12 months ?? costly ?? Everytime you contact them with a problem it is always your pc yet everything was working fine before they did the upgrades !!! i am getting really tired of Topaz and lack of help and support to sort issues out.

1 Like

Which one do you use?

2 Likes

Almost any new graphics card bought in the last 12 months should be ok for TPAI, even at the relatively low-end of the scale (I’m still using a GTX 1060 6GB that originally came out in 2016). Which I suppose then begs the question, what did you get?

1 Like

Recommended graphics card:

Graphics Card (NVIDIA)|NVIDIA RTX 20 series or higher
Graphics Card (AMD)|AMD Radeon 5000 series or higher

Is there any effort been made for the Text Preservation thing? Oh my gosh, it is bad! The result is very very very bad. It clearly doesn’t work.

The result was good before. What happen?

This should be a Severity 1 Defect! I was a QA before.

4 Likes

This version update did not load properly from the app. It wants to load modules every time I open, and it doesn’t function properly. Tried to repair by uploading on Topaz download site, but keeps quitting saying I have no connectivity - which it total bs. I don’t have any other program I update with connectivity issues except Topaz! So now I’m not happy