Topaz Sharpen AI update from my previous post.

[Locton]

I thought I would post separately to my previous post about my issue with Topaz Sharpen AI. I have had a reply from Topaz and anyone who has a D500 should get the same issue I had, it may also happen with other models, although there seems to be no comment on Nikon Cafe if this has happened.

The following is what Topaz has said:

Yazi Saradest (Topaz) Mar 16, 2021, 11:07 AM CDT Hi Jeremy, Thanks for getting back to us and sending in the images. We've been able to reproduce the color change with the file you sent and this seems to be a result of a conflict between our RAW processing libraries and your particular camera file/type. This may change over time as our RAW libraries get updated again in the future. For such RAW conflicts from any camera type, we would recommend converting those files to .TIFF prior to importing into our software. This will allow you to have a data-rich file for editing (like a raw file) with few, if any, of the common conflicts associated with raw editing. I hope this helps clarify, but please let me know if you have any questions, or if I can help with anything else, Jeremy. Yazan Saradest Product Support Specialist M-F: 09:00am-05:30pm CST Become A Topaz Labs Beta Tester:

 

[drr1531]

Not sure that I see an advantage to opening a RAW file with AI Sharpen. It only does one thing and that typically comes later in the workflow. So why use it to open RAW files at all? Seems like a non-issue. Can't imagine why Topaz even put that function into the product. But thanks for the info.

 

[Mike Buckley]

AI Sharpen. It only does one thing and that typically comes later in the workflow.

Though I mentioned it in another thread, I'll also mention it here: Topaz Sharpen AI is fundamentally a substitute for capture sharpening that has gone awry. So, my thinking is that it should come at the very beginning of the post-processing workflow. That thinking is also consistent with Topaz's information at their website.

 

[Growltiger]

Though I mentioned it in another thread, I'll also mention it here: Topaz Sharpen AI is fundamentally a substitute for capture sharpening that has gone awry. So, my thinking is that it should come at the very beginning of the post-processing workflow. That thinking is also consistent with Topaz's information at their website.

I two-thirds agree. I think that since Topaz isn't a proper raw processing program, the sharpening should come as early as possible after raw development.

So do very simple raw processing with a good program, don't sharpen much if at all. Output as TIF (best) or JPG. But denoising should be before sharpening so either do that when raw processing, or if serious denoising is needed use Topaz Denoise.

Only then do Topaz Sharpen.

After that do all the rest of the pixel based postprocessing that you want.

 

[drr1531]

Though I mentioned it in another thread, I'll also mention it here: Topaz Sharpen AI is fundamentally a substitute for capture sharpening that has gone awry. So, my thinking is that it should come at the very beginning of the post-processing workflow. That thinking is also consistent with Topaz's information at their website.

That's an interesting take on it. Please share the info from the Topaz website that suggests sharpening at the very beginning of PP. When I look I'm unable to find such a suggestion in any of the text on the Sharpen AI product page. If indeed somewhere they are suggesting using AI Sharpen as step one in RAW conversion then they are internally confused. Per the screen capture below the workflow in the video tutorial that they have linked shows doing basic RAW processing and converting to TIFF prior to using Sharpen AI. Interestingly the video that they link is nearly a year old and for v 2.0.x which is not the most recent. However since that is the tutorial they're still using I assume they haven't changed their suggested workflow.

Subscribe to see EXIF info for this image (if available)

In regard to "capture sharpening" most of the references I've read promote sharpening during RAW conversion in order to offset image softness caused by anti-aliasing filters on camera sensors. Most(maybe all?) Nikon DSLRs of the past couple of iterations do not have anti-aliasing filters. I eliminated capture sharpening from my workflow years ago with no apparent ill effect. My default workflow in LR has the sharpening slider set at zero.

 

[Mike Buckley]

Please share the info from the Topaz website that suggests sharpening at the very beginning of PP.

When reviewing the web page shown below, keep in mind that I never need the first step because I always begin with a raw file and I rarely need the second step because I rarely have noise in my photos. So, the third step that they recommend is typically at the beginning of my workflow.

https://topazlabs.com/the-ultimate-workflow-for-topaz-labs-ai-software/

As for eliminating capture sharpening, neither I nor my wife (I post-process all of her images) has owned a camera that doesn't have an anti-aliasing filter including my Nikon Z6.

More important, the one time I didn't put Sharpen AI at the beginning of my workflow, the results were terrible. This was when I was using the trial license. The results were so bad I planned not to license the product. Then I realized that I had applied Sharpen AI "on top" of my typical workflow sharpening. Once I resolved that issue, I purchased a license.

 

[Mike Buckley]

I two-thirds agree. I think that since Topaz isn't a proper raw processing program, the sharpening should come as early as possible after raw development.

We actually completely agree about that. My take on that is that I convert the raw image file, in my case using Adobe Camera Raw, and before making any adjustments I then use Sharpen AI. Then I make all the other adjustments needed. One reason for that is that the highlights can be increased when using Sharpen AI (not for any fault of Sharpen AI) so much that it makes sense to me to wait to adjust the tone curve until Sharpen AI has been used. When not using Sharpen AI, I always think of the tone curve as my first basic adjustment, though there are a few very minor adjustments I automatically apply to all images using a batch process to save time before adjusting the curve.

The few times I've used Sharpen AI could probably be counted on one hand and I've only needed it for my drop art photos. Even so, the process I described worked like a charm every time.

 

[Mike Buckley]

Per the screen capture below

That workflow chart makes no sense to me. That's because, by definition, workflow inherently takes place one step at a time. Yet the chart indicates Sharpen AI and Denoise AI taking place at the same time, which is impossible to make happen. Similarly, I don't know how to interpret the chart displaying use of Sharpen AI and all the other adjustments to the left on the same level. I think it's an ineffective chart that doesn't work well on its own without explanation and clarification of the information it's intending to provide.

This reminds me of the time I was unsuccessfully trying to get CamRanger to work, so I called the customer support folks. The guy I got, who I later learned is one of the top people in the company, kept using a particular term that made absolutely no sense to me in the context of how their software works. Then he finally explained that the term he and the company use is actually an incorrect use of that term. Once he mentioned the accurate term, everything made perfect sense and I was able to get everything to work quickly and easily. I asked why he and his company use the inaccurate term and he responded that most photographers don't even know the definition of the accurate term; his contention was that using the accurate term would only confuse most photographers.

 

[GonzoBernelli]

Last year I was a guest star in the Topaz dog and pony show. AI Sharpen did absolutely nothing after a software update. My file was a 16 bit TIF processed by the then current Capture One because it originated from a Phase One digital back. I sent Yazi the TIF via dropbox. And his response was the same one quoted above - it works here!. We went back and forth several times until I finally gave up on Topaz’s obfuscation and demanded they return the money I paid for Sharpen and Denoise (similar D&P show). I deleted everything with “Topaz” in its name as a final and permanent fix for their software.

 

[TonyW]

Probably teaching a few grannies how to...... but.....
I believe the godfather of sharpening was the late Bruce Fraser. AFAIK he came up with the 'ideal' sharpening workflow i.e. multipass Capture, Creative and output sharpening

Capture Sharpening is applied as the first step in the image-editing process and just aims to restore any sharpness that was lost in the capture process. With or without AA filter all Bayer sensor captures will benefit to some degree to the first round of sharpening i.e. capture sharpening if only to restore the information lost in the demosaicing process. This allows the image data to respond well to subsequent sharpening such as creative and finally output sharpening. The ideal way is to use deconvolution algorithms that target and try to reverse the loss of sharpness, but do not add light and dark halos such as USM etc. ACR and Lightroom have a fairly rudimentary but useful slider that uses both USM type and deconvolution, the Detail slider. There are better deconvolution sharpening algorithms in other raw editors such as Rawtherapee and Darktable enable many iterations of the algorithms

With all that said, I am trying to understand how the Topaz product with (by all accounts!) a relatively poor raw engine can really do deconvolution sharpening for captures then as a second-round apply selective/creative sharpen/desharpen and finally output sharpening which by nature is global.

 

[Mike Buckley]

I am trying to understand how the Topaz product with (by all accounts!) a relatively poor raw engine can really do deconvolution sharpening for captures then as a second-round apply selective/creative sharpen/desharpen and finally output sharpening which by nature is global.

The only help I can provide is that when using ACR rather than Sharpen AI to convert the raw image file, the workflow you described worked for me the few times I tried it.

 

[Lucky Duck]

I deleted everything with “Topaz” in its name as a final and permanent fix for their software.

Permanent is unlikely since they sometimes undo the good with more bad via their updates. Having said that, I do find that Sharpen AI is viable for some situations. I would not consider it a capture sharpener, at least as I practice capture sharpening. That role now falls to DxO. I export chosen frames (as dng) from PL4 (using only optical corrections and noise reduction) for further processing elsewhere. The optical corrections include lens sharpness, which works fine as capture sharpening for me. The noise reduction is DeepPrime which is state of the art IMHO. Sharpen AI comes into play for frames that need something that no other software can provide. Sometimes it works, sometimes it doesn't—which amounts to a coin toss—and so far, I can't predict how that coin will land.

 

[Lucky Duck]

Here's an example of the results of my still evolving workflow, using DxO only for capture sharpening and NR. The first image is as rendered straight out of C1. I've adjusted color and exposure but sharpening and noise reduction were left at defaults, which I generally find to be appropriate or at least good starting points. The second image is the dng exported from PL4 and rendered in C1 using the same settings as previous but with NR and sharpening set to 0. The last two are these same images as processed by Sharpen AI, auto values. All are 200% screen shots.

Note that if you copy and paste them into Photoshop, the dng images will not overlay the C1 images perfectly due to PL4's lens distortion corrections.

D810, AF-S NIKKOR 70-200mm f/4G ED VR, ISO 1600 for those without access to EXIF.

Subscribe to see EXIF info for this image (if available)

Subscribe to see EXIF info for this image (if available)

Subscribe to see EXIF info for this image (if available)

Subscribe to see EXIF info for this image (if available)

 

[drr1531]

That workflow chart makes no sense to me...I think it's an ineffective chart that doesn't work well on its own without explanation and clarification of the information it's intending to provide.

In all fairness the chart was used as part of a video tutorial which does(I assume) provide clarification. Also I guess I'm guilty of one of those mental blocks(I forget which) in the thread a while back and assumed that flow charts are intuitively obvious to everyone I agree it could be better. If there was a decision block directly beneath the TIFF block that says "NR necessary?" and yes/no arrows to Denoise AI/Sharpen AI respectively it would be more clear.

This reminds me of the time I was unsuccessfully trying to get CamRanger to work...his contention was that using the accurate term would only confuse most photographers.

Must have been a term like "100 percent crop"

I believe the godfather of sharpening was the late Bruce Fraser. AFAIK he came up with the 'ideal' sharpening workflow i.e. multipass Capture, Creative and output sharpening...Capture Sharpening is applied as the first step in the image-editing process... The ideal way is to use deconvolution algorithms that target and try to reverse the loss of sharpness, but do not add light and dark halos such as USM etc.

I believe that the original multi-step sharpening process mentioned was developed before deconvolution and/or "AI" sharpening algorithms were available. Or at least before they were available to the average photographer. All of the articles I recall reading about the multi-step process employed USM.

With all that said, I am trying to understand how the Topaz product... can really do deconvolution sharpening for captures then as a second-round apply selective/creative sharpen/desharpen and finally output sharpening which by nature is global.

There is an argument to be made that deconvolution and/or AI sharpening processes have rendered multi step sharpening obsolete/unnecessary.

 

[Growltiger]

I hope you don't mind, Lucky Duck but I took the second of your images and gave it a very quick go. All I did was first remove the noise (DeNoise AI) using no sharpening, and then sharpened (Sharpen AI). I think the result looks pretty good, although starting from the original would have been better, since starting with the 200% version has introduced pixellation artefacts:

Subscribe to see EXIF info for this image (if available)

 

[TonyW]

I believe that the original multi-step sharpening process mentioned was developed before deconvolution and/or "AI" sharpening algorithms were available. Or at least before they were available to the average photographer. All of the articles I recall reading about the multi-step process employed USM.

It was and for capture sharpening you would normally expect to apply the smallest radius available at a very low level hardly noticeable as it is only intended to remove the slight softening from the demosaic process or a little more if an AA filter in the pipeline. For Deconvolution to work correctly an estimate of PSF needs to be made and the number of iterations to achieve the correct levels. Adobe has only partially implemented this but it seems that they are making small tweaks under the hood with each update including so-called AI .

There is an argument to be made that deconvolution and/or AI sharpening processes have rendered multi step sharpening obsolete/unnecessary.

Not sure who is arguing that and for now, I stand to be convinced after admittedly only a short play. I am not saying that it does not work well, clearly, it does and makes many people happy?

It seems to do several things, Sharpen for general, Stabilise shake correction and Focus that may help with lens blur. It does not appear to offer any attempt at deconvolution (as used in many scientific areas including forensic work and recovering detail from the Hubble telescope) or output based sharpening. The latter requires the application to know what image size, PPI, what viewing condition, what type of medium etc. and can vary tremendously e.g. 720ppi image to print will require a much larger radius setting than a 360ppi image.

So as a replacement to standard sharpening using USM, Smart Sharpen, or High Pass I can appreciate there may be something to gain, image dependent of course. But the indications are that it may not be optimal over a good capture sharpening regime and a final global sharpening for output device

 

[Mike Buckley]

Must have been a term like "100 percent crop"

No, it wasn't that term. I don't even remember the term. But I do know that if he had used the term, 100% crop, that would have been my signal to hang up.

 

[Palouse]

No, it wasn't that term. I don't even remember the term. But I do know that if he had used the term, 100% crop, that would have been my signal to hang up.

Oh, 100% crap?

 

[Ann_JS]

Input Sharpening (via the Details panel in ACR) is best kept to the minimum necessary; and Noise reduction likewise. Any reduction of Noise also blurs the definition of detail so I try to reduce Luminance Noise by the least possible amount (preferably zero) while reducing Colour Noise separately.

Bruce Fraser (who I knew: he and I would often discuss and argue about all sorts of technical issues on Internet forums late into the night!) was among the Pixel Geniuses who developed his ideas and writings about Sharpening into the PG Toolbox Scripts (which I mentioned in another thread here).
https://www.nikoncafe.com/threads/e...raws-new-super-resolution.326897/post-4150245

The Scripts originally had to be purchased but they are now being offered as a free download.

The installation of the Scripts on Mojave and later versions of OSX is a little complicated but the excellence of the Scripts makes the installation worth doing.

I always turn-off all Output Sharpening in ACR itself and use the appropriate PG Script for my intended Output to make the top-most Layers in Ps as the final step before Output.

 

[Mike Buckley]

There is an argument to be made that deconvolution and/or AI sharpening processes have rendered multi step sharpening obsolete/unnecessary.

There is always an argument for something. The question is whether the argument is valid. Seriously, the one step of sharpening that probably will never be obsolete at least in theory is the output sharpening, which will vary from output to output.