Jono Slack Review: Leica SL2-S

[Chaemono]

Advertisement (gone after registration)

vor einer Stunde schrieb Daedalus2000:

And to show an example of what I mean, if we take the picture that @Chaemono posted above and take a crop you will see that it will not clean all areas, and for some it  will keep the noise or reduce it less, like the highlights in the crop below. I am not sure how the algo works, but I am guessing the model underneath it tries to identify the areas where the image has information (details) and therefore should not be denoised too much and areas where it does not have information/detail and need to be smoothed. Sometimes it gets it wrong, observe how smooth the darker areas are below and how noisy the highlights are...

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

Another example from one of the iso 12500 pics from @Chaemono, observe how the noise between the branches is still high (and on the bokeh balls) but other blue areas have been smoothed a lot.

There are trade-offs. At some point AI will become so good, it’ll be able to distinguish and adjust even these areas. It just has to look at thousands of Leica bokeh balls first. 😂

Edited December 30, 2020 by Chaemono

[SrMi]

1 hour ago, Daedalus2000 said:

And to show an example of what I mean, if we take the picture that @Chaemono posted above and take a crop you will see that it will not clean all areas, and for some it  will keep the noise or reduce it less, like the highlights in the crop below. I am not sure how the algo works, but I am guessing the model underneath it tries to identify the areas where the image has information (details) and therefore should not be denoised too much and areas where it does not have information/detail and need to be smoothed. Sometimes it gets it wrong, observe how smooth the darker areas are below and how noisy the highlights are...

<snip>

Another example from one of the iso 12500 pics from @Chaemono, observe how the noise between the branches is still high (and on the bokeh balls) but other blue areas have been smoothed a lot.

<snip>

I have also observed inconsistencies in Topaz DeNoise AI: flat surfaces having varying amount of noise reduction. Therefore, I prefer doing it manually (LrC or PS), unless I am in a rush :).

Edited December 30, 2020 by SrMi

[Daedalus2000]

2 hours ago, Chaemono said:

There are trade-offs. At some point AI will become so good, it’ll be able to distinguish and adjust even these areas. It just has to look at thousands of Leica bokeh balls first. 😂

True 😂

[Chaemono]

vor 9 Stunden schrieb SrMi:

I have also observed inconsistencies in Topaz DeNoise AI: flat surfaces having varying amount of noise reduction. Therefore, I prefer doing it manually (LrC or PS), unless I am in a rush :).

You can’t gat the little guy with the wings and the candle to look like this from an SL2 ISO 3200 file unless you apply some kind of AI. One of the most overlooked aspects of Topaz DeNoise is that it also reconstructs the objects in focus.

[LocalHero1953]

I use DeNoise AI and Sharpen AI almost interchangeably - they both seem to apply noise reduction and sharpening, but with different emphasis. Both have automatic settings which I use initially, check the result, and then play around with the settings to see I can make it better - often I can. I agree both can introduce artefacts, but even so I find them both preferable to the blanket noise suppression of Lightroom which turns faces into polished plastic. Photoshop can manage it better than LR  because you can use masks for selective noise reduction (in principle you can in LR as well, but the noise reduction in local adjustment tools is simplistic).

I have been exploring DeNoise and Sharpen on various crappy images from my past: high ISO low light M9 shots, scanned film, jpgs, tiffs. The Topaz tools are undoubtedly best on modern raw images IMO - it doesn't seem to understand film and jpg noise well enough to make much difference.

I have a desktop PC with a high end graphics card, which makes it easy to use these tools to preview changes in real time; I can imagine that on a lower spec device they could be a real pain to use in anything but full auto mode, and even that would be slow.  

Edit: Unlike @Chaemono I have always reduced noise reduction, sharpening, clarity and texture to zero before editing in Topaz. I will try his method of applying some NR in Lightroom first, to see if it helps with analogue and old jpg noise.

Edited December 31, 2020 by LocalHero1953

[Daedalus2000]

2 hours ago, Chaemono said:

You can’t gat the little guy with the wings and the candle to look like this from an SL2 ISO 3200 file unless you apply some kind of AI. One of the most overlooked aspects of Topaz DeNoise is that it also reconstructs the objects in focus.

 

1 hour ago, LocalHero1953 said:

I use DeNoise AI and Sharpen AI almost interchangeably - they both seem to apply noise reduction and sharpening, but with different emphasis. Both have automatic settings which I use initially, check the result, and then play around with the settings to see I can make it better - often I can. I agree both can introduce artefacts, but even so I find them both preferable to the blanket noise suppression of Lightroom which turns faces into polished plastic. Photoshop can manage it better than LR  because you can use masks for selective noise reduction (in principle you can in LR as well, but the noise reduction in local adjustment tools is simplistic).

I have been exploring DeNoise and Sharpen on various crappy images from my past: high ISO low light M9 shots, scanned film, jpgs, tiffs. The Topaz tools are undoubtedly best on modern raw images IMO - it doesn't seem to understand film and jpg noise well enough to make much difference.

I have a desktop PC with a high end graphics card, which makes it easy to use these tools to preview changes in real time; I can imagine that on a lower spec device they could be a real pain to use in anything but full auto mode, and even that would be slow.  

Edit: Unlike @Chaemono I have always reduced noise reduction, sharpening, clarity and texture to zero before editing in Topaz. I will try his method of applying some NR in Lightroom first, to see if it helps with analogue and old jpg noise.

I have found the PRIME noise reduction in DxO Photolab to be very good and recently I started using the DeepPRIME in DxO Photolab 4 which also uses AI. It seems to work well but I have not done extensive work with it and unfortunately it depends on the software supporting your camera's raw files. 

[Photoworks]

Advertisement (gone after registration)

2 hours ago, Daedalus2000 said:

 

I have found the PRIME noise reduction in DxO Photolab to be very good and recently I started using the DeepPRIME in DxO Photolab 4 which also uses AI. It seems to work well but I have not done extensive work with it and unfortunately it depends on the software supporting your camera's raw files. 

DxO does not support leica files from sl2 or newer.

Topaz really need to be used with blend in original. Otherwise it looks so fake

[SrMi]

On 12/30/2020 at 8:36 AM, Dr. G said:

This is good information.  I never understood why Topaz recommended DeNoise as the first step before any processing, including level adjustments (at least they used to).  

AI based noised reductions are apparently trained on files coming from cameras, not from post-processed files. This means that they may be more effective before any post-processing is applied.