Removing "red lines" from Objects (Ringing? What is it?)

[Platos]

I have a picture here (uploaded). Look on his shoulder, there is a red line. How can i remove that? So i mean what is this called? Ringing? I tried HQDering and it did not help anything. So i guess its not ringing. Also i tired DeHalo_Alpha, but it does also not work. Anyone has an idea?

[poisondeathray]

Looks like slight color channel misalignment

one way to treat it

mergechroma(awarpsharp2(depth=6))

[Platos]

i tried it with aWarpSharp2 in Hybrid with Planes "chroma only" and a thick on "chroma" and it did not work (vapoursynth).

I tried it with chromashiftSP in hybrid. It goes away, when i use Y=2, but then it also appear red lines on other objects (in the other direction). When i use Y=0.5 it does somehow not mess up too much but also fix the red line above a bit. But it is no real fix, more a workaround.

I also Tried FixChromaBleeding with Shift 0/1 and Strength 1.0 and different Threshold from 0.3-12. More or less it is the same as ChromashiftSP with Y=1. Because you can only increase shift in 1 and not 0.1 there is no comparison with Y=0.5


Do you have any other idea, which i can use in hybrid?

[poisondeathray]

I don't use hybrid (selur can help) but the equivalent in vapoursynth using awarpsharp2 port would look like
https://github.com/dubhater/vapoursynth-awarpsharp2

I used mergechroma helper function in muvsfunc
https://github.com/WolframRhodium/muvsfunc

.
import muvsfunc as muv
.
.
.
## assuming YUV clip
# clip = your source

aws = core.warp.AWarpSharp2(clip, depth=6)
mc = muv.MergeChroma(clip, aws)
mc.set_output()

To apply filter to specific sections, you can use trim to divide into segments, apply filters selectively and splice

[Platos]

Ahh, so it is basically not possible to fix the red line "globally" without handmade segments? Because i mean this is only an example, so if i want to fix all red lines, i would have to hand-fix thousands of frames by hand ? Would it work, if you do not use trim function or would it then mess up on other objects like on my try?

If someone knows about hybrid: Can i use muvsfunc in hybrid? Is there any Option to "globally" remove this "chromatic abberation" without messing up other details? Because "handmade" frame by frame fixing is to much time consuming. Then i would have a year, till im done.

[Selur]

Can i use muvsfunc in hybrid?

sure, you can use anything in a custom section

Is there any Option to "globally" remove this "chromatic abberation" without messing up other details?

No, since nobody knows a method to identify "chromatic abberation" and stuff you can't identify usually can't be removed automatically,...

Because "handmade" frame by frame fixing is to much time consuming.

If you want suggestions for video, share a video clip.

About the mergechroma: Hybrid also has 'Merge'-options which allow you to weight YUV channels.

[Attachment 88060 - Click to enlarge]
Using Y:0,Y:1,V:1 will replace the chroma planes of the input with the filtered output.
Thus you would get code like:

clipMerge = core.warp.AWarpSharp2(clip=clipMerge, blur=2, depth=6)
clip = core.std.Merge(clip,clipMerge,[0, 1, 1]) # merge filtered and unfiltered by plane

So, no need to use muvsfunc.MergeChroma, but like I wrote you can use it using a custom section.
(Hybrid already has muvsfunc in the vsscripts section.)

Cu Selur

[Selur]

SSIQ does seem to help: https://imgsli.com/NDAzMTk2

[poisondeathray]

Ahh, so it is basically not possible to fix the red line "globally" without handmade segments? Because i mean this is only an example, so if i want to fix all red lines, i would have to hand-fix thousands of frames by hand ? Would it work, if you do not use trim function or would it then mess up on other objects like on my try?

That relatively low depth setting for awarpsharp2 is unlikely to mess up other frames ; ie. you could probably use it globally without trim. It's easy to preview and check before you encode, and you can change the settings

Conversely, shifting the chroma 1 direction on every frame is ALWAYS going to mess up "good" frames. The difference is the chroma alignment is guided by the luma for those default awarpsharp2 settings, so the adjustment is per frame guided by the higher resolution 1920x1080 Y channel. The chroma channels are only 960x540 for 4:2:0

[Platos]

Sry for the late answer guys!

Ok, so then when i understand correctly, the best suggestions till now, which does mess up the rest least, is:

SSIQ (DeRainbow stuff in hybrid)
And aWarpSharp2 (i would like to use Hybrid with Vapoursynth).

Which settings in hybrid would work for aWarpSharp2? Because i did not found out a solution, which does work. I see no reduction with that.
And which settings did you use selur with SSIQ ? Default ?

Ok, i did upload a 5s long video (unfiltered): https://uploadnow.io/f/qlJgQ7S
And here already filtered with Degrain and a bit Sharpening (as in the picture): https://uploadnow.io/f/zwkZ9jW

And about intendet effect: On another version it has not that strong red lines, so i think it is not thaaat intendet. But the other version is worse on other things so...

[Selur]

That source if HDR. Are you converting HDR to SDR (if you do, what settings are you using) or do keep the HDR data as is?
(this might be a bug in the HDR2SDR filter used)

If you are using a HDR2SDR filter, try:

# color adjustment using ToneMap (Placebo)
clip = core.placebo.Tonemap(clip=clip, src_csp=1, dst_csp=0, src_min=0.0050, src_max=1000.0000, dst_min=0.2023, dst_max=203.0000, dynamic_peak_detection=1, percentile=99.99, gamut_mapping=9, tone_mapping_param=0.500, contrast_recovery=0.30)
# adjusting color space from YUV444P16 to YUV420P8 for vsSSIQ
clip = core.resize.Bicubic(clip=clip, format=vs.YUV420P8, range_s="limited", dither_type="error_diffusion")
# rainbow removal using SSIQ
clip = core.ssiq.SSIQ(clip=clip, interlaced=0)

with the latest dev (updated libplacebo to the current version from https://github.com/sgt0/vs-placebo)

If you keep the HDR meta data, the issue is probably in the meta data,..