Color banding and noise

[cungybuns]

I see color banding and noise in Kodi, MPC-HE, PotPlayer, and pretty much any player except mpv. I do not use any sort of post-processing in them (including mpv). Before anyone asks, yes, I have read over video output and color space extensively. The colors are correct but these ugly artifacts are present and there's no way they're actually present in the source. I've tried different displays and the problem is still present. You may say "if it ain't broke don't fix it" in regards to this problem not appearing in mpv, but I want to use a madVR-compatible player for tone mapping HDR to SDR, and mpv is not one of those.

I've included images from mpv and Kodi in the Google Drive link below to show what I'm referring to. I use a VA panel for watching movies and shows, but I have trouble seeing the artifacts on my laptop's IPS panel as the images are dark and IPS panels suffer from backlight bleed. Therefore, you may have a bit of trouble seeing the banding and noise depending on what monitor you're using to view the images.

https://drive.google.com/drive/folders/1...sp=sharing

[Emulgator]

Your link is incomplete.

[cungybuns]

Your link is incomplete.

My apologies. Nobody had replied for awhile prior to your response, so I just took the link down, assumed the thread would become buried in the depths of Doom9, and stopped checking for replies. I've created a new, working link and shall vigilantly await your response...

https://drive.google.com/drive/folders/18rk1iqwxwzK8NyYXe4K4c9eNzWEt4RGY?usp=sharing

[foxyshadis]

mpv dithers by default, I'm not sure any other players do. You can achieve comparable shots just by turning on dithering, but how each does it is definitely not standardized. I'm pretty sure mpv injects a tiny bit of noise in the process to "boost" the dither for many scenes that are already flattened in the source; MadVR will do something similar if you set it to 6-bit output, which isn't ideal, but beats flat washed-out surfaces on bad 8-bit video. MPDN had a fantastic renderscript to inject a tiny bit of noise via shader, but I haven't been able to get MPDN to run in years.

[pirlouy]

Sorry I can't help, but if foxyshadis is in the right direction, by default, I think dithering is disabled in mpv (dither-depth=no) and enabled in madVR ("Ordered Dithering").
In mpv you can enable it by using "profile=gpu-hq".

[cungybuns]

mpv dithers by default, I'm not sure any other players do. You can achieve comparable shots just by turning on dithering, but how each does it is definitely not standardized. I'm pretty sure mpv injects a tiny bit of noise in the process to "boost" the dither for many scenes that are already flattened in the source; MadVR will do something similar if you set it to 6-bit output, which isn't ideal, but beats flat washed-out surfaces on bad 8-bit video. MPDN had a fantastic renderscript to inject a tiny bit of noise via shader, but I haven't been able to get MPDN to run in years.


If I'm understanding you correctly, you're saying the artifacts I'm seeing are part of the source? I'm willing to accept that conclusion, but one of my reservations about doing so is that these are Blu-ray rips I made myself, and they're of high-budget productions like Better Call Saul and Westworld. Surely, there wouldn't be so many artifacts in the Blu-rays of these shows?

Furthermore, I did what pirluoy suggested and experimented with the --dither-depth and --profile settings in the mpv configuration file. --dither-depth is set to no by default, but when I removed --profile=gpu-hq, I got an image similar to that I've seen in other players—one accompanied by artifacts.

You also mentioned mpv injecting some noise into the video by default. I looked through the manual and found an option by the name of --deband-grain. The manual stated that its set to 48 by default as it "significantly helps cover up remaining quantization artifacts." I set it to 0 in my configuration file yet found no discernible difference.

[huhn]

dither it self is noise.
it trades banding for noise.

[cungybuns]

dither it self is noise.
it trades banding for noise.

Yes, but there are two separate options: --dither-depth, which is turned off by default, and --deband-grain, which is set to 48 by default from a range of 0–4096. I had both of them turned off and there was still no banding when --profile was set to gpu-hq. However, when I removed --profile=gpu-hq, banding appeared even when --deband-grain was at its default value of 48. This leads me to believe that the lack of artifacts in mpv, whether it's through post-processing or not, is related to --profile being set to gpu-hq.

[huhn]

debanding is unrelated to dithering. they are often used together because but dithering doesn't need a deband filter.
sorry this is going no where without a sample.

[pirlouy]

Once again, I can't really help but be careful with profile=gpu-hq since it also enables debanding.
To see which options are activated when using gpu-hq profile:
mpv.com --show-profile=gpu-hq

In my case, I have this:
scale=spline36
cscale=spline36
dscale=mitchell
dither-depth=auto
correct-downscaling=yes
linear-downscaling=yes
sigmoid-upscaling=yes
deband=yes

So if you really want to compare things, you'd better use these options individually.

[cungybuns]

debanding is unrelated to dithering. they are often used together because but dithering doesn't need a deband filter.
sorry this is going no where without a sample.

I included samples in a Google Drive link in a previous reply. Here's the link: https://drive.google.com/drive/folders/18rk1iqwxwzK8NyYXe4K4c9eNzWEt4RGY

[Emulgator]

What the others said.
The Kodi screenshots are 3840x1902 32bit .png, the mpv are 1920x1080 24bit .png.
Not even the same frames...Do we compare apples to oranges ?
Asking for samples might also mean:
Do we get to see source video to be able to compare various players' renderings against the decoded picture?
With blu-ray it will be 8-bit anyway and yes, there may be visible steps encoded around low Y and around neutral U,V,
Happens with professional Blu-ray and UHD disc productions, my drawer has a good part of these.
Visible Blocks in UHD black letterboxes, "zer"graded shadows from major labels, anything...
Which dithering will nicely cover, and it might be sitting hidden within one player's chain and not the other's.

[cungybuns]

What the others said.
The Kodi screenshots are 3840x1902 32bit .png, the mpv are 1920x1080 24bit .png.
Not even the same frames...Do we compare apples to oranges ?
Asking for samples might also mean:
Do we get to see source video to be able to compare various players' renderings against the decoded picture?
With blu-ray it will be 8-bit anyway and yes, there may be visible steps encoded around low Y and around neutral U,V,
Happens with professional Blu-ray and UHD disc productions, my drawer has a good part of these.
Visible Blocks in UHD black letterboxes, "zer"graded shadows from major labels, anything...
Which dithering will nicely cover, and it might be sitting hidden within one player's chain and not the other's.

The difference in color depths and resolutions was an oversight on my part, I apologize. The Google Drive link below contains three different video samples that I think are representative of the problem: two from Better Call Saul and one from Westworld. I made sure that no re-encoding was performed.

I'm a video purist, so I'd rather see the artifacts if they're actually in the source rather than use post-processing to remove them. Also, what are these "'zer'graded" shadows" you speak of?

https://drive.google.com/drive/folders/1VZ82KbYWEZfpFHA6xsYjZRHpYRENWpoX?usp=sharing

[Asmodian]

However, when I removed --profile=gpu-hq, banding appeared

be careful with profile=gpu-hq since it also enables debanding.

Sounds like this is it. Do you still hope for more investigation?

I believe the artifacts are in the source, I certainly notice some issues in the source with your samples above.

[cungybuns]

Sounds like this is it. Do you still hope for more investigation?

I believe the artifacts are in the source, I certainly notice some issues in the source with your samples above.

I don't need anymore investigation. You and pirlouy seem to have solved the issue. I thank both of you and all others who responded.