[SOLVED] Can't get smooth gradients in blacks

Lately I shot on dark background. The cyclorama was a little bit aged and stomped. When trying to clean things up I get severe banding as shown (slightly exaggerated with curves to be more noticeable).
screenshot6

So,

  1. Exporting with background untouched looks as expected (spotted but no banding).
  2. After smoothing the background (tried dt, RT, GIMP, Krita, all in 32-bit mode and different methods) the background looks okay while in editor yet gets banded after export (PNG-16 looks almost as bad as JPEG)
  3. Adding noise to darks with grain module makes the banding less pronounced.

My initial guess was it simply lacks bit depth, but it looks quite decent untouched, even if exported in JPEG, and PNG-16 didn’t help a lot.
The question is if someone has ever bumped onto this issue and if so, how was that solved? Also – how do you usually get smooth gradient in dark background?
UPD: Of course, I can provide a RAW if need be.

Forgive me, but I have no idea what this means.

1 Like

Of course I will)) The cyclorama is special studio background where the wall and the floor are painted the same color and the joint is rounded to not create noticeable shadows. And that construction, being in heavy use in studio, has scratches, foot prints, and so on, visible on an image.

Have you tried the dithering module?

Well, it’s the same. Any retouch adds banding in darks.

Some additional observations.

  1. Retouching with fill tool (color inpainting) does not add colored banding (well, maybe just a little bit). Smoothing with the blur tool does. (Dithering ON).
  2. Exposure with feathered mask, graduated density do not.

So I can get acceptable results (but not ideal) combinig those tools. As well as prove PNG-16 is capable to bear these gradients. I’m afraid there’s some glitch in blending math (or in exporting). As soon as I’m sure it’s not my mistake I shell probably file a bug.

EDIT: Exposure with mask does add banding…

Early in the development of my hack software, I experienced something similar in the display rendition. I determined that I was applying the display profile after I’d converted the internal working image to 8-bit for display, so I re-ordered my display pipeline and it went away.

That’s not something you can mess with as a user, so, just saying…

This is obviously not a display issue, because I see no banding until I export the image. However, geeqie may be involved… Will investigate this tomorrow…

Oh, I misread your initial post.

Post your raw, we’ll make sure the input image is not at fault. I still think it has to do with a ICC transform, though…

Which one? Input or output? DT has input ICC profile for Pentax a little bit strange, I often dislike the colors it produces. I normally use alternative profile downloaded from I-dont-remember-where. I got few of them, none is completely satisfactory. I’d like it to be a no-brainer, I mean, one-fits-all-sizes, freeing me from fiddling with input profiles now and then. So I’m looking forward adding DNG profiling to DT, assuming that 2 points are better than one)))
However, I think you’re right. Having changed the profile to dt builtin seems to solve the issue. (I’d never guessed that input profile may be involved))) I think I need to play with it some more before marking in solved, but at least I switched off grain and post-sharpening diffuse-or-sharpen and still see no obvious banding or color blotches.
Meanwhile here goes the RAW. CC-BY-NC for forum members. Have fun!
20220115_0027.PEF (40.4 MB)

Few more words on dithering module. TBH I never saw any improving from it on my photos, that’s why I never use it and didn’t even remember it was there)) Maybe it is useful for cheaper cameras with small sensors, but starting with Nikon APS-C and now with Pentax FF I never got banding. Until now)))
Anyway, it did not help this time (yet grain did, in a sence).
Later update is in post 43.

Oookey. Finally I got it. Seemingly))
That was the combination of two issues.

  1. Input color profile I used provoked banding somehow. Damn!
  2. PNG export is broken. Exporting in JPEG quality 98 gives no banding, while export in 16-bit PNG produces banding and colored stripes even in previously fully desaturated areas. Which is complete nonsence.
    And for completeness,
  3. Adding some grain in dark areas (via parametric mask) helps further (ISO 50, strengh 10%, midtone bias 50%). Unfortunately it’s still in LAB. BTW with dithering still see no difference.

Eventually I have smooth transitions even in JPEG. Geeqie is redeemed from being a suspect.
Later update: Eventually Geeqie is to blame for lying to me (and eog, too). PNGs produced by dt are fine, viewers show them incorrectly.

1 Like

You’ll only get that ‘broken PNG export’ investigated and fixed, if you provide input (= raw + processing instructions) to trigger the issue.

Interesting, I was trying unsuccessfully to refer to the output profiles, which are for the display and export destinations. I don’t know dt code, but for the input profile to be involved indicates a tortuous path…

I’d agree, doesn’t make sense to me either. 16-bit png should give plenty of bit-space to gradiate…

doing such just mitigates a poor situation upstream, which needs to be addressed first.

Yes, the devs will need to see your tool chain to isolate the problem area.

Oh, your subject needs an amp. I just can’t get past that cord just winding around to the floor, unplugged… :crazy_face:

3 Likes

You’re a true geek, if there ever was one! :rofl:

That’s magic wifi transmission :yum:
Actually, there was a combo in a studio, but it was a concert simulation, so we decided not to clutter the shot. (I just moved the subject and did not pin down the cord end showing up)))

1 Like

Sorry, I didn’t see you already posted the raw. Could you please also post the sidecar (xmp), so we start with the same processing steps?

You mean here or on github? Or both? I just prepared the stuff for uploading where it’s more pronounced (actually, the very shot I’ve been struggling with).

I meant here; but if you decide to raise an issue, you should also attach it there. Maybe it’d make sense to check here if there really is a bug to be fixed, before you file a report.