RawTherapee Sharpening Question

[xpatUSA]

RawTherapee (RT) provides two methods of sharpening, USM or RL (but Gaussian) de-convolution. Not asking which is better nor asking how they work. The question is "does use of the one exclude use of the other?". RT sharpening is enabled by a single check-box and the parameters for each method are shown on-screen and are also listed in the sidecar file.

I ask because some folk (including Bruce Fraser, RIP) find that several stages of sharpening can be beneficial and, in many editors, this can be done. RT, on the other hand, is a 'one-shot' affair, hence this post.

I'm hoping to be told that I could have a large radius, low amount USM combined with a small radius larger amount de-conv. which, with a tiny bit of capture sharpening in Sigma PhotoPro, would constitute the much-vaunted 3-step process pushed by Cambridge-in-Colour.

So there it is - in the sidecar file - does RT act on every tag in the sharpening section i.e. both USM and RL de-conv?

Thanks,

[Bart_van_der_Wolf]

I'm hoping to be told that I could have a large radius, low amount USM combined with a small radius larger amount de-conv. which, with a tiny bit of capture sharpening in Sigma PhotoPro, would constitute the much-vaunted 3-step process pushed by Cambridge-in-Colour.

So there it is - in the sidecar file - does RT act on every tag in the sharpening section i.e. both USM and RL de-conv?

Hi Ted,

As far as I can see, it's either USM or RL (consistent with the preview), but not both. RawTherapee does allow to read the TIFF output file though, so if you apply RL Capture sharpening on Raw conversion you can then, in a new run on the TIFF output, add USM Creative sharpening. However, the object of Creative sharpening is often a localized adjustment, so it's more logically applied by a dedicated photo-editor application with selections and masks.

Cheers,
Bart

[xpatUSA]

Thanks Bart, it's as I feared ;-)

Since I'm only using TIFFs as input to RawTherapee (it doesn't convert X3F files) there are opportunities to apply "capture sharpening" in some converters that I use, e.g., Sigma ProPhoto, dcraw, etc.

Lately however, adventuring has been in the world of IR - where everything is twice as blurry, ho ho - and my current preference for conversion is RawDigger's RGB TIFF export. It does no processing at all but does come up with a nicely tinted image :-)

Thanks,

[Fine_Art]

I agree it is an over-site for them to make it either/or. I don't really like their R/L implementation. You are better off in a dedicated program like focus magic (I use images plus). If you have decent equipment the output files have enough fine detail that you do not need much sharpening.

I get the impression they are proud of their USM implementation, with good control of edges only. It is very good as far as USM can go.

[xpatUSA]

. . . .You are better off in a dedicated program like focus magic (I use images plus). If you have decent equipment the output files have enough fine detail that you do not need much sharpening . . . .

Quite so.

However my situation is perhaps rare in that, although I am intensely interested in the technical aspects of photography, I am certainly no photographer myself. Thus I am quite content to shoot in low resolution (shock, horror) 1134x756px on on my early Sigma DSLRs and view the results on my un-calibrated (shock, horror) 1280x1024 NEC monitor and never print anything.

Low res shots can be very sharp indeed but, for IR work, much less so as I said above - hence my increased interest in sharpening methods.

cheers,

[ablankertz]

What you're looking for -aka LCE or HIRLOAM sharpening- is hidden as "Local Contrast" in the Shadows/Highlights section of the _Exposure tab.

[MichaelEzra]

Ted, the USM and RL sharpening are mutually exclusive in RT. This has been inherited from the original RT design, but I don't really see a reason why it has to stay that way.
The USM employs a very unique threshold adjustment where effect can be modulated across the tonal range, to control its falloff strength in shadows and highlights.
This provides an excellent control to avoid over-sharpened results.

Additionally, there are few other sharpening tools available in the detail tab.
I would certainly recommend Contrast by Detail Levels (CBDL). I would not recommend combining the finest detail adjustment in CBDL and the USM.
If USM is used, the larger scale detail should be sharpened via CBDL.

In the new version of RawTherapee CBDL gets a modulation by hue to either target or bypass skintones.

[xpatUSA]

Thank you Michael and @ablankertz for your good advice.

I'll look at all those methods.

Might even read the manual sometime

cheers,

[Isaac]

Finally I've installed RawTherapee and with several choices for every processing step I'm definitely going to be reading the manual.

[ablankertz]

Finally I've installed RawTherapee and with several choices for every processing step I'm definitely going to be reading the manual.

I think that RT's weakness is that there are too many options, which might be because it's open source. It's strength is its demosaicing algorithms, not only Amaze , but also the two that are optimized for noisy shots, lmmse and igv. Nothing comes close to their ability to deal with noise- you have to see it for yourself.

[Isaac]

I think that RT's weakness is that there are too many options...

So far, I haven't experienced that as a weakness.

For example, CIECAM02 might be "best" but that doesn't matter until I understand what advantages CIECAM02 offers me. Until then I can ignore CIECAM02 completely. Lab seems to offer more understandable advantages, and I can benefit from that option immediately.

Six months from now I might be using a different set of options -- but that doesn't hinder my use of RawTherapee now.

[ppmax2]

I think that RT's weakness is that there are too many options, which might be because it's open source. It's strength is its demosaicing algorithms, not only Amaze , but also the two that are optimized for noisy shots, lmmse and igv. Nothing comes close to their ability to deal with noise- you have to see it for yourself.

I would love to see your workflow and what settings you are using for demosaicing and NR--I feel pretty good about my skills in other tools but am struggling to get decent results with RT. I have a particularly challenging shot that I use as a benchmark with Aperture, LR, C1, and RT...and C1 blows everything away. In fact that is really the only reason I bought C1...to deal with those tough shots. Rt is the type of app I would love to love, but I haven't really seen much good come out of it yet.

Thx
Pp

[Fine_Art]

I would love to see your workflow and what settings you are using for demosaicing and NR--I feel pretty good about my skills in other tools but am struggling to get decent results with RT. I have a particularly challenging shot that I use as a benchmark with Aperture, LR, C1, and RT...and C1 blows everything away. In fact that is really the only reason I bought C1...to deal with those tough shots. Rt is the type of app I would love to love, but I haven't really seen much good come out of it yet.

Thx
Pp

I would bet money RT beats LR and C1 in detail and noise reduction with little effort. If you put in more effort you might be able to beat them using the LAB controls for color. The USM is probably as good as it gets without neighborhood adaptation (Images plus) the richardson lucy is ok, the defaults are wrong. It should be 100 cycles 65 strength instead of 75, 65 radius as well. Use microcontrast 3x3. Try contrast by detail (a particularly useless name) 1.2 for the 3 finest unless you use USM. If you use recovery the color propagation is quite good. Output your image using 200% LANCZOS. Compare that to your "benchmark".

Contrast and saturation should be via lab not the normal controls.

[Fine_Art]

Lets do a comparison on the Imaging Resource standard RAW for the Sony A7R
Here is the file:
http://www.imaging-resource.com/camera-reviews/sony/a7r/AA7RhSLI00100NR0.ARW.HTM

Attached are my basic processing parameters in RT.Output gamut is AdobeRGB1998. It's a free program, try it.
Output at 200% using lanzos. Output with your own parameters from LR or C1.

Compare side by side.

If someone can show better results from another program they have "proof" they can beat a low effort RT conversion. On to round 2.

Link to 200% output. Argb 1998 gamut.This is +2 ev.
http://www.megafileupload.com/en/file/549882/AA7RhSLI00100NR0-crop-jpg.html


Here is a crop 200% of their fabric circle.
http://www.megafileupload.com/en/file/549884/AA7RhSLI00100NR0-crop2-jpg.html


Man, their setup is dusty.

[ppmax2]

Thanks for the replies Fine_Art--much appreciated.

First, I'm not trying to start a holy-war...just saying that I haven't yet discovered the secret sauce with RT yet   I want to like and know how to use this tool...

Regarding your suggestions for RL:
Those are much better than the defaults--thanks for those.

Regarding the A7R studio shot:
I'm not really after a comparison of these tools with ideal lighting...I'm happy with Aperture for those types. I should have clarified in my OP: I have a few after-sunset shots where I exposed for dramatic skies (5DM3, ISO 800, 1/160, f5, 70-200mm f4 IS USM) and the shadow areas are underexposed and need help with noise  and detail.

I've got an early morning tomorrow but will post my CR2 and a big JPG of what I developed with C1 when I get home tomorrow night. I hope you don't think I've put the challenge back on you...I would just love to know *how* to tickle RT so that I can get a result like that I've created with C1. I'll download the files you linked to and will see what I can do with those in C1 and Aperture.

Regarding:
>>Rt is the type of app I would love to love, but I haven't really seen much good come out of it yet.

This came out wrong: I should have said that most of the images I've seen in the Play Raw competitions have been pretty underwhelming. This is not to say that the tool isn't capable of great results.

Thanks again and have a good one--
PP

[xpatUSA]

Thanks for the replies Fine_Art--much appreciated.

First, I'm not trying to start a holy-war...just saying that I haven't yet discovered the secret sauce with RT yet   I want to like and know how to use this tool...

This topic being one I started sometime back, I feel moved to comment:

I have stuck with RawTherapee (RT) since the original post and have grown accustomed to it's novel user interface. The tools are there aplenty (some have said "too many") to do just about anything 'better' than other, less comprehensive editors. It takes awhile to become really au fait but the time spent is well worth it - if image quality is your goal. If, on the other hand, producing tens or hundreds of 'OK' jpegs for the web is your need, then stuff like LR is perhaps your bag.

cheers,

[Fine_Art]

If there is a raw converter that can beat RT I expect it would be DxO based on their lens correction database. They make a very smooth file, especially at high ISO with their prime NR. It is a very good starting place for deconvolution. It has 2 relative weaknesses:
-it tries to do too much under the hood taking control away for the user (good and bad)
-they charge double for full frame cameras which is a blatant money grab.

You can do your own lens corrections in RT, if you have a way of testing. Most people don't have tools to do it easily so they don't bother.

I have seen mazing on C1 images. LR always looks soft. If recent builds have caught up to the detail level in AMAZE or VNG it would be news. I honestly want to see their output at 200% on the standard Imaging Resource RAW.

[ppmax2]

Hello Fine_Art--

I had some time to do some developing with C1 and RT using the Sony raw file you posted. Here are some files for you to review.

In C1 I did the following:
1. Set the exposure curve to linear, then bumped exposure by 1.25. No curves or levels changes.
2. Bumped "Structure" in the Clarity brick to 21 (using the Neutral method).
3. Set white balance by sampling the 2nd to last grey color patch (2nd from bottom right). White balance reports as 6394 with tint -3.8.
4. Applied global sharpening (Unsharp Mask) Amount=168, Radius=1.1, Threshold=0.

C1 (crop, 100% scale)
http://ppmax.duckdns.org/public.php?service=files&t=b5fbedb89a2353a2f3da664219d9708d

C1 (full size image, 100% scale)
http://ppmax.duckdns.org/public.php?service=files&t=5116c2b1395a5be0ea3f81e42f2fcad6

C1 -> TIF -> RT (crop, 200% scale)
http://ppmax.duckdns.org/public.php?service=files&t=b77934910010f7648f05ac201be73200

C1 -> TIF -> RT (full size image, 200% scale)
http://ppmax.duckdns.org/public.php?service=files&t=585c8131d646ff2a14d7e82931f07ede

In RT I did the following:
1. Bumped exposure up to 1...until the clipped highlight indicator showed just a few sparkles on the 2nd and third yarns to the right of the folded napkins. (I've gotta say I like the more saturated color rendition vs. C1).
2. Added RL sharpening using the values you posted previously (Radius=.65, Amount=.65, Damping=20 (default), Iterations=100). I also tried Unsharp Mask using the values set in C1 above...but thought RL looked better.
3. Enabled Microcontrast with 3x3 matrix; Quantity=20 (default) and Uniformity=50 (default).
4. Enabled Contrast by Detail Levels using the suggestions you made above (top 3 sliders set to 1.2; all others default values).
5. Set white balance by sampling the same color chip as above (2nd from bottom right). WB reports as 5409 with tint at 1.111 and Blue/Red Equalizer at 1.000.

RT (full size image, 100% scale)
http://ppmax.duckdns.org/public.php?service=files&t=2d8ea68ee2f8b2d9ace2b81957d91315
RT Output PP3
http://ppmax.duckdns.org/public.php?service=files&t=e58523b504ad6c228dd6fb32d1be65a2

RT (full size image, 200% scale)
http://ppmax.duckdns.org/public.php?service=files&t=8d9e789e1b37de47adbfd8601d543220
RT Output PP3
http://ppmax.duckdns.org/public.php?service=files&t=273c675446718ea0e7f8648caa945b1a

All images should be Adobe RGB(1998).


I think RT's color rendition is more punchy while C1 is more muted. I imagine that the real thing is somewhere in between...FWIW I didn't do any color tweaks in C1 or RT.

I think RT handles the highlights better. The highlights on the metal parts of the brushes looks great, while C1 seems to render less "bloom" in the highlight areas. This could be my fault...since I was probably more careful with exposure in RT (avoiding clipped highlights).

I like C1's rendering of the text better--it's more black...sharpness seems to be a wash.

I like C1's rendering of the napkin texture better; in RT the dark red napkin above the pink napkin looks a little mushy to me...this could be due to over saturation and could probably be fixed.

Would love to hear your feedback on these.


Next, I'll post my after-sunset CR2's and 2 renders from C1. I'd love to see what you can do with the noise on these.


Best,
PP

[Fine_Art]

The C1 upscaled in RT looks pretty good. A hair less definition. I agree the blacks look nice and dark. On mine I set exposure +2ev without moving the black point.

The point of the 200% was that I expected the RT lanczos to beat the other raw converters unless they were using the same. It is not a small point, for printing we often need lots of pixels.

I see a faint halo on all high contrast edges.

The cloth does look very good, maybe better than RT. The texture looks very natural. Interestingly the texture of the fuzzy cloth in the cup is poorly defined. Maybe it is a difference of finest detail. RT does the fuzzy cloth and the dust specks well. the thread of the cloth probably needs adjustment in contrast by detail to match the very good C1.

I like the color of C1 better. I didn't have a color profile for the A7r so it is set on Sony.

[Fine_Art]

I put the laptop away to compare your images on a real screen.

A couple of things stand out. The C1 has way more contrast, I would say artificially high.That is probably why the thread in the cloth circle is so well defined. Look at it on the thread of the white cloth under the cup. In RT it looks like a soft cloth, in C1 it looks harsh like plastic. Also look at the tips of the crayons. In RT the waxy satin look is there, in C1 it again looks artificial. If you look at the tan cloth above the crayons C1 does it well while RT shows color aliasing. To fix that go to the demosaicing tab. set the remove false color pre-processing to 2 or 4. The highlights on the labels all look natural in RT. In C1 they look plastic. Maybe toning down the contrast will fix that. The bristles on the brushes look far more natural in RT. If you look at the thread top right, the contrast makes them look like nylon in C1. In RT they look softer like cotton.

I agree on the colors (this is a better screen than before) the RT looks better even though the color checker shows they are over saturated. C1 has more accurate color when looking at the color checker. Strangely the red, orange, pink, etc look like real colors in RT while they look strange in comparison in C1. Its hard to fathom that, you would think accurate color would look better.