Differences in the output sharpness of JPG when processed with DT or LR

I think CA correction is too aggressive, it desaturates not only the edges. If you pixel-peep, you’ll see the roof has lost the red tinge, even though it was probably not CA:

Zoomed out, it’s not bad at all, though, you just get a little bit of desaturation.

I think the sharpness preset of diffuse or sharpen does a decent job:

Combined with a bit of DoS dehazing (see XMP):


PK320007.DNG.xmp (23.6 KB)

(The sidecar is based on the one by @Donatzsky, I took his tone equalizer and filmic settings, added the denoising discussed in a previous post and replaced the sharpening steps.)

For reference, I tend to think there is something going on with optical corrections.
Below are some screenshots before / after in the DXO Pure RAW preview:

One way to measure sharpness is conversion to the frequency domain, voila:

Shows that LR (right) sharpened a bit more than DT (left) …

2 Likes

In the LR settings used by the OP I wonder if the shadow, highlights and dehaze adjustments are having an impact on the sharpness of this image. I know that shadow and highlights adjustment can affect sharpness in DT from personal experience and the dehaze setting even though very small would have some effect on sharpness in LR.

LR has a tendency to be heavy handed in some defaults settings because that is what the general user likes. On the other hand DT is less heavy handed and puts the burden of decision making on the user. This is both a blessing and a curse in DT because people want great results straight out of the box but DT wants to leave the creative decisions up to the users.

In my posts here I have increased the sharpness trying to match LR, but in reality I felt that degree of sharpness was neither necessary or desirable given the subject was shot from a distance over water and there is a natural softening effect of atmospheric haze present. But this all comes down to personal taste.

4 Likes

There are. Chromatic aberrations, as I mentioned further up.
Not sure what DXO does exactly, but those results have not just been corrected, but also sharpened.

Well said … and “sharpness” can mean many things to many men.

If I recall correctly, Adobe’s ACR applies default sharpening (25?) and it would not surprise me if LR does the same …

Yes, it does. Moreover, the same value of 25 means different strength for different cameras.

1 Like

Just an observation: The bell tower crop shows a notable difference in highlights. DT highlights are set way higher. Those areas cover the details with a “bloom” effect - see the tower’s clock.

We may have been trying to describe the same sort of thing… I have not done any thorough testing to see what might be any sort of display artifact vs what actually comes out of the export but it was also what I was referring to about ART in that it seems to avoid that while sharpening…but you can use local contrast to knock that down and really improve the results making it more like what ART and RT come up with… I think anyway… I tried to demonstrate this in this short clip… adding the LC tweal will tame the blowout…

Sorry for the late response, but it is stressful times due to an upcoming conference. First to all I’ll like to thanks to all of you who took your time to give an answer to my question. Said that, some things to say:

  1. Errata: just to clarify. I said previously that didn’t applied a sharpness to the image generated with LR, which is not true (in any case, everyone correctly assumed that it had been applied.). I though that I had deselected that module. Here are the sharpening settings:

imagen

And here is a comparison of the JPG with and without sharpness mask.


(100% crop. Left: LR JPG with sharpness. Right: LR JPG without sharpness)


(200% crop. Left: LR JPG with sharpness. Right: LR JPG without sharpness)

  1. Similar topic: @Andy_Bulka last year made a very similar question. So here are additional answer to basically the same question. Here the link to the topic https://discuss.pixls.us/t/how-to-get-darktable-sharpness-to-match-lightroom/35125

  2. Taking into considerations all the different suggestions and ideas I think I finally achieved, at least to me, a fine result that match the results from LR. Here is the image comparing both JPG in a 200% crop. In summary, I sed the “diffuse or sharpen” module with “sharpen demosaicing: no AA filter” and twist the sharpness (at the bottom) to 25% (still remains to recover the colors lost by the CA module).


(200% crop. Left: LR JPG. Right: DT JPG).

Here is the XPM file if any one of you would to explore how I got the result.
PK320007.DNG.xmp (20.1 KB)

Thanks to @Donatzsky for the detailed asnwers and the video to CA, also to @kofa for the explanation of denoised module which I use to get this result. Thanks @Terry for the nice ideas (also played with the sharpen module but I never got a result that fullfil what I was looking for), and thanks to @priort for been the first to help me :smile:

I’ve never explored the DorS module before, so there is a lot potential to explore there.

Note: if someone would like to compare in full detailed both JPG here they are:
LR jpg:

DT jpg:

3 Likes

This is a truly great module. I would recommend at first using the presets that are available. I take snapshots to compare the results and decide which provides the best sharpening. I use lens deblur a lot as an extra instance in addition to the sharpen demosaicing: no AA filter which is my initial sharpening applied to nearly all images. The developer made this fairly technical video explaining how it works if you are interested. https://www.youtube.com/watch?v=DREdq7guNz4

I am only now as a result of your post here beginning to explore the option of creating some of my own presets. It is a very confusing module for the average person, but the presets save the day. In DT version 4.8 there will be a new preset called local contrast: fine and this has a very nice ability of bringing out fine detail in the images.

1 Like

Thanks! I compared them in the frequency domain (Fourier analysis) … the LR has slightly more detail texture than the DT … I can post the FFT’s if anyone is interested.

1 Like

I took a look at the xmp…consider reseting the sharpness slider at least to begin with and instead try 3 maybe 4 but no more iterations of the preset…I think you will find that more than sharp enough and perhaps a better result than with sharpness… you can tweak further with a small opacity change if it goes too far in a particular image or a tweak of the details threshold in the masking tab to target more precisely the application of the module in the case you get a bit of noise amplification…

1 Like

While playing around I noticed this

If you are using snapshots to make comparisons it will be slow but if this is verified snapshots might only be accurate if you enable the preview HQR preview. You can see what I mean here… I have taken a snapshot and then enabled it to compare it against the same image…they should match… and they do if you enable HQR in the preview which is not an option in the current release version but it could be related to this possibly?? If anyone else notices this you could also comment on the issue I started on Github…

I am using the master branch and self compile for WIndows… it was happening on 4.7 +1237 and I recompiled tonight and noticed the same thing… I saw it with OPENCL on and off and with a different image…

@priort I’ve never used snapshot before, but on Sunday I had the same problem. I didn’t give it much thought, but now that you brought it up it could me something.

@xpatUSA I’m interested in the the FFT’s function :smiley:

I’ve put the plots up on my site:

http://kronometric.org/phot/temp/DT%20LR%20FFTs/

For each scene, one plot is a G’MIC FFT; the other is the same FFT with the GIMP color>‘Threshold’ option set to 0.5 for ease of comparison.

1 Like

Hello Todd,
please, con you tell me how can it be enabled?
Thank you

If you use 4.8 or above it’s an icon at the bottom of the screen . I’m on my phone … It’s near the color assessment icon if I recall

1 Like

I’m very late to this thread so apologies if this has been linked already but this thread - Settings for fine details (compared to other tools) - #35 by nosle
had some useful input on this subject.

1 Like