Detail Refinement in the Develop Persona: extraordinarily good results... but how???

[KLE-France]

Hi to all.

The basics: Windows 10, Affinity Photo 1.10.5 (but this has been true since my first-ever version of AP), usually 20 mp-ish images out of a Nikon D7500.

Here's what one gets searching for site:https://forum.affinity.serif.com/ "detail refinement".

Therein is this post where James Ritson (of Affinity Photo's how-to videos (excellent work James!)) says that "detail refinement is indeed an unsharp mask filter," and adds, speaking of the strange use of percentages therein, that "(they are) simply 1:1 with the pixel radius (so 100% = 100px)." There are also other finds in the search result where various people say that Detail Refinement is basically a USM filter, that the USM filter in the Photo Persona does the same thing, etc. etc...

But it doesn't. At least in my case. The Detail Refinement function gives way, way way better results than any of the other various methods of sharpening in AP. But it does not behave like a USM filter does.

Going through the various search results and the AP help, I cannot find any convincing explanation for how it works.

Because in my experience at least, it does not work like a USM filter.

Try this: Open a raw photo in AP, get your exposure somewhere acceptable, then activate Detail Refinement, set Radius to 30% (or even 100%) and amount to 100%, and use Noise Reduction to tame any over-sharpening of tiny detail (or not!). So, according to James's description above, you've got a 30 pixel radius and 100% sharpening. That should, especially in my 20, 21 mpx images, give horrendous halos.

But it doesn't:

 

But hey, it's supposed to be a USM filter, so let's deactivate Detail refinement, develop the image (opening it in the Photo Persona), then apply an equivalent USM, i.e., 30 px radius, 100% factor (4) and a bit of threshold to keep it from over-sharpening small detail:

 

 

Yikes! No, that's not behaving at all like Detail Refinement.

Detail Refinement seems to be doing something in the vague concept of capture sharpening. And it's splendid. I love it. You can push it as far as you want, Detail Refinement won't halo; even 100% for both Radius and Amount won't cause haloing--and sometimes even works fine for certain images.

But how? I am hoping that maybe a developer can step in to clarify what exactly Detail Refinement does.

Sure, I don't need to know the answer to that question; I can just take advantage of this unmatched sharpening tool. But, enquiring minds and all that...

By the way and to all: you can also use it post Photo Persona: once you've got your image where you want it do a Merge Visible and send that layer over to the Develop Persona for Detail Refinement. You won't be disappointed.

Discuss discuss and thanks in advance to anyone who can shed light on this AP element.

 

 

[Old Bruce]

14 minutes ago, Kerwin said:

But hey, it's supposed to be a USM filter, so let's deactivate Detail refinement, develop the image (opening it in the Photo Persona), then apply an equivalent USM, i.e., 30 px radius, 100% factor (4) and a bit of threshold to keep it from over-sharpening small detail:

Why is 4 assumed to be 100%? And why "a bit" for the Threshold?

[Lisbon]

Hi Kerwin.

4 hours ago, Kerwin said:

Therein is this post where James Ritson... says that "detail refinement is indeed an unsharp mask filter,"... various people say that Detail Refinement is basically a USM filter, that the USM filter in the Photo Persona does the same thing, etc. etc...

But it doesn't. At least in my case.

Interesting. In the image that I tested I got very similar results with this settings.

Detail Refinement (Develop Persona)
Radius: 100% / Amount: 100%

USM (Photo Persona)
Radius: 4px / Factor: 2 / Threshold: 0

4 hours ago, Kerwin said:

Because in my experience at least, it does not work like a USM filter.

USM can be more aggressive. Notice that on my previous settings Detail Refinement is maxed out.

In your first screenshot you have "Detail Refinement" and "Noise Reduction" active. They have almost like an opposite effect. One tends to soften. The other sharpens. Therefore, for a fair comparison between USM and Detail Refinement, uncheck "Noise Reduction".

4 hours ago, Kerwin said:

But hey, it's supposed to be a USM filter, so let's deactivate Detail refinement, develop the image (opening it in the Photo Persona), then apply an equivalent USM, i.e., 30 px radius, 100% factor (4) and a bit of threshold to keep it from over-sharpening small detail:

Different units of measurement. Notice that one comes in percentage and USM in pixels.
A radius of 30% is different from 30px.

After a bit of trial and error, Radius: 30% / Amount: 100% should correspond more like Radius: 3px / Factor: 2 / Threshold: 0

Again this is where the values I got in my test image.

[KLE-France]

@Old Bruce

"Why is 4 assumed to be 100%?"

Because 4 is the maximum factor applicable on the USM filter (thus "100%", like I did for Amount in Detail Refinement). 'Not me naming these things! 

"And why "a bit" for the Threshold?"

Because a applied "a bit" of noise reduction to the same ends (avoid oversharpening small details) in the Develop Persona. 'Not performing a carefully-designed scientific experiment here! And see below.

 

@Lisbon

"In the image that I tested I got very similar results with [these] settings: Detail Refinement (Develop Persona) Radius: 100% / Amount: 100%;  USM (Photo Persona) Radius: 4px / Factor: 2 / Threshold: 0"

Yes, I'm not contending that you can't get similar results with the USM filter. But now, crank up that USM filter's radius to 100 px (100% like in Detail Refinement) and its Factor to 4 (100% like in Detail Refinement) and see what you get.

What's interesting to me, is that in Detail Refinement, Which, according to James and others, "is indeed an unsharp mask filter," you can do exactly what you did and have perfectly acceptable, que dis-je, very good sharpening. I have never seen a USM filter that can be pushed to its maximums and give good results; doing so in any USM filter in any photo retouching program (well, those I've known) will give garish, clownish results. But doing so in Detail Refinement doesn't. Which suggest that its underlying calculations are different from those of a "conventional" USM filter, or in any case that something is different with it (continue reading).

"In your first screenshot you have "Detail Refinement" and "Noise Reduction" active. They have almost like an opposite effect. One tends to soften. The other sharpens."

Yes, I know. The noise reduction tactic for Detail Refinement is nothing more than that: just a tactic. It happens to limit fine detail sharpening when using Detail Refinement, in a "visually similar" way to Threshold in the USM filter (yes I know they're different).

"Different units of measurement. Notice that one comes in percentage and USM in pixels. A radius of 30% is different from 30px."

Not according to James Ritson, who, in his response here (same link as above) says: "(...) detail refinement is indeed an unsharp mask filter. The use of percentage rather than pixels for the radius is basically an oversight: at one point, there may have been further designs (e.g. adaptive sharpening), but as it stands, the percentage is simply 1:1 with the pixel radius (so 100% = 100px)."

If 100% equals 100 pixels in Detail Refinement then 30% equals 30 pixels. James's comment is by far the most qualified explanation that I've been able to find. If someone has a better one, or a newer one, from a qualified person, please post a link! That's exactly what I'm looking for.

But that "percentages equals pixels" doesn't seem to make sense, which is my point: Detail Refinement does not behave like the USM filter does. Or, if it does, its indications are extremely misleading when placed in comparison to AP's own USM filter. You finished with this: "After a bit of trial and error, Radius: 30% / Amount: 100% should correspond more like Radius: 3px / Factor: 2 / Threshold: 0." Quite possibly! And if so, Detail Refinement's indications should be: Radius 0 to 10 pixels (or to 4 pixels? See your first equivalency test) and: Amount 0 to 2 factors (currently no Threshold adjustment there). I Note too that there is a precedent for limited effects in the Develop Persona. For example, Saturation increase is limited to 50% in the Develop Persona; If one wants really garish saturation, one has to jump over to the Photo Persona.

So yes, what I'm hoping for is not how to-s but an explanation on the very different (and very good!) behavior of the Detail Refinement function in the Develop Persona.

Thanks to both for your input!

[Lisbon]

5 hours ago, Kerwin said:

But now, crank up that USM filter's radius to 100 px (100% like in Detail Refinement) ... and see what you get.

In the example I tested, 100% did not match 100px. This is verifiable both by looking at the differences in the images as well as the USM itself.
Since the radius of the USM goes up to 100px one might think it corresponds to 100% but in fact the value of 100px can easily be changed to 200px by simply entering the value manually.
And now, does 100% correspond to 200px?

5 hours ago, Kerwin said:

I have never seen a USM filter that can be pushed to its maximums and give good results

Your right. And the reason for this is actually interesting.
USM is used primarily for sharpening.
When you sharpen, you are increasing the contrast at the edges. The size of the edge is controlled by the radius.
So if you increase the radius, the area to be sharpened becomes larger, and at some point it also starts to affect the overall contrast.
Depending on the initial amount of contrast, the image can get really ugly at this point.

5 hours ago, Kerwin said:

Not according to James Ritson, who, in his response here (same link as above) says: "(...) detail refinement is indeed an unsharp mask filter. The use of percentage rather than pixels for the radius is basically an oversight: at one point, there may have been further designs (e.g. adaptive sharpening), but as it stands, the percentage is simply 1:1 with the pixel radius (so 100% = 100px)."

I'm sure he said it for a good reason but to be honest I don't understand his point.

5 hours ago, Kerwin said:

Detail Refinement does not behave like the USM filter does.

It seems to me that Detail Refinement is just a softer version of USM. That's why you don't get those "clownish results" that you mentioned before.

5 hours ago, Kerwin said:

You finished with this: "After a bit of trial and error, Radius: 30% / Amount: 100% should correspond more like Radius: 3px / Factor: 2 / Threshold: 0." Quite possibly! And if so, Detail Refinement's indications should be: Radius 0 to 10 pixels (or to 4 pixels? See your first equivalency test) and: Amount 0 to 2 factors (currently no Threshold adjustment there).

Hard question.
From the values I've deduced, this doesn't look like a linear relationship, making it hard to know.