Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion. Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules. |
5th February 2024, 12:58 | #1 | Link |
_
Join Date: May 2008
Location: France
Posts: 692
|
Judder with several media players
I'd like some feedback with this sample pls:
https://0x0.st/HkfY.1.mkv - Put your monitor in 60 Hz - Play this 30 fps sample in full screen with your preferred application - Focus on the moving upper bar - Then compare with Windows default apps: "Movie & TV", or "Media Player" or "Photos" in full screen (I don't know why there are so much apps for the same thing but whatever) In my case, with Windows default apps, ok there are sometimes dropped (or repeated) frames but it's way smoother than my usual apps. I've tried this with mpv (default config or with D3D11 Hwdec) and MPC-BE with MPC Video Renderer and both are choppy compared to Movie & TV app. Is it the same to you ? Don't know if it can change something, but I'm on Windows 11, Full HD monitor, Nvidia GTX 1060. |
5th February 2024, 13:55 | #3 | Link |
_
Join Date: May 2008
Location: France
Posts: 692
|
What about this 60 FPS sample, with 120 Hz monitor (hoping you have one) ?
https://0x0.st/Hkfn.2.mkv You really mean there's no difference in terms of motion ? |
5th February 2024, 20:39 | #4 | Link |
Registered User
Join Date: Oct 2012
Posts: 7,925
|
i only tested sample hold with refresh based there will be a difference but not between player.
as long as the file doesn't break in playback the cadance will be perfect. photo sometimes hicks up. if you do full screen with this file there is a good chance it breakes. may it is the 1 sec audio in a 10 min video don't know. |
6th February 2024, 00:09 | #5 | Link |
_
Join Date: May 2008
Location: France
Posts: 692
|
I didn't believe you at first, but man, telling huhn is wrong seemed wrong. I've tried to close all programs, and I found the culprit, it was an app which caps Windows timer/clock resolution. After stopping this app, now I have the same behavior between Movie& App and mpv or MPC Video Renderer.
Despite this, I still don't understand why I have hiccups sometimes. Let's say with the 60 fps video on 120 Hz monitor (OK, it's 119,98), it will be smooth for 30 seconds, then there will be hiccups for 3~5 seconds, then smooth again. In MPC Video Renderer, there's a line which seems to show there's no problem but I can see these hiccups during 3 seconds. I though it could be another program, but all programs are closed, there's only the video player (no apps in systray, nor service, I don't have antivirus). I would understand if there was a dropped frame from time to time because it's not exactly 120 Hz, but no hiccups for 3 seconds. I don't understand what could cause this. I've tried exclusive full screen mode, but it's the same behavior. And 120fps video on 120 Hz is working flawlessly. 60 fps video on 60 Hz, perfect too. Infuriating ! |
6th February 2024, 00:26 | #6 | Link |
Registered User
Join Date: Oct 2012
Posts: 7,925
|
there is a lot going on so yes this can go wrong.
when i make it go full screen at UHD i get 1 fps at best. could be the windows wddm not doing it's job properly and the video renderer may not even know about that. or the odd audio stream don't know. if you care: https://forum.doom9.org/showthread.php?t=185247 |
6th February 2024, 00:36 | #7 | Link |
_
Join Date: May 2008
Location: France
Posts: 692
|
Ah yes, I saw this thread, but it's way more advanced for me. So these hiccups are unavoidable ?
FWIW, hiccups are even more frequent at 240 Hz. With Windows 7, it was possible to disable Aero. OK, you could have tearing, but at least no problems linked to desktop composition. I guess I could use things like resampling in mpv (or madVR smooth motion), but it will just hide these hiccups and adds some blur. |
6th February 2024, 00:48 | #8 | Link |
Registered User
Join Date: Oct 2012
Posts: 7,925
|
in the past there where ways to avoid them now no none i know of that will work on all system.
for me mpcVR has hick ups on my 240 hz screen with exclusive 4ms judder some times. madVR does just fine but it is an AMD system they are not known with any frame time issues unlike nvidia where frame time issue are common and where some workaround are just applied generally. which was disabling mpo forcing vsync and setting queue to 1-3. on my HTPC UHD 120hz with 4060 i use the default queue of 8 with normal driver setting mpo and it is working fine. because it doesn't affect all in the past fullscreen exclusive and overlay rendering where very good reliable ways to fix this issue. both are dead now. overlay has been discontinued on a driver level and exclusive is just pretended by the OSD. intel technically still has overlay which maybe false for the dGPU and only a thing iGPU have. edit: tscale oversample or smoothmotion will not hide these hickups they maybe even more visible with that because these algorithm rely on that just as normal playback. |
6th February 2024, 12:19 | #9 | Link |
_
Join Date: May 2008
Location: France
Posts: 692
|
"tscale oversample" really does a difference, but at cost of GPU and CPU (ok, not that much but noticeable) and surely a bit of blur.
I don't know if you know mpv that much (I don't know what you chose as your default media player since you know a lot...), but I can tell you how to check the difference. In your mpv.conf, you can use this: video-sync=display-resample interpolation=yes tscale=oversample And in your input.conf, you put: F10 cycle-values video-sync display-resample audio (or other hotkey) Then you put your monitor in 240/280 Hz, and you should notice a difference with/without interpolation. Last edited by pirlouy; 6th February 2024 at 12:38. |
6th February 2024, 14:39 | #10 | Link |
Registered User
Join Date: Oct 2012
Posts: 7,925
|
i don't know much about mpv but the settings you just showed is something i know about.
and no for this black white bar test it doesn't matter the screen ghost so hard on it own it doesn't matter i can try my oled with BFI later but for this gaming screen it does not matter. but as you said i never verified if oversample which according to the manuel is a SM clone is actually an SM clone. at 144 "audio" setting is not "working" as in it is not smooth that supposed to be the reclock approach isn't it? |
6th February 2024, 15:44 | #11 | Link |
_
Join Date: May 2008
Location: France
Posts: 692
|
"tscale=oversample" is supposed to be a clone of madVR but it is not exactly the same, since audio can be resampled, whereas madshi explained he didn't want to touch audio.
Some links if you're interested: https://github.com/mpv-player/mpv/wi...n#display-sync https://github.com/mpv-player/mpv/wiki/Interpolation True smooth motion like madVR has been requested but it was not understood by developers: https://github.com/mpv-player/mpv/issues/7674 There's no reclock clone "video-sync=audio" is just the default setting: frames are repeated/dropped according to audio like in all default media players (I think). Last edited by pirlouy; 6th February 2024 at 22:58. |
6th February 2024, 22:30 | #12 | Link | |||
Registered User
Join Date: Oct 2012
Posts: 7,925
|
Quote:
there is 0 point in touching audio when smooth motion is active. the algorithm is so simple in theory it's laughable easy. the written one is most likely not easy not at all. you take the frametimes correct them by the audio video drift and blending them by how much time they are of the vsync in linear light. example (very simple numbers): frame time 21 ms corrected to 20 ms by audio dirft refrshrate is 13 ms. frames are now listed as F1 F2 and refresh are R1 R2 R1 is F1 because 20 is more then 13 rest 7 R2 is 7/13 of F1 and 6/13 of F2 rest 14 R3 is F2 rest 1 R4 is 1/13 of F2 and 12/13 of F3 rest 8 in the end 23p source at 120 hz is played by exactly 48 different frames a sec the rest is repeats even if it targets 240 it is still 48 the combination is always blended frames normal frame repeat normal frame. in a realistic use way. madVR SM can do a 120 hz file and show it at 23p but that looks like shit. Quote:
display-resample is according to your link reclock. Quote:
|
|||
6th February 2024, 23:20 | #13 | Link |
_
Join Date: May 2008
Location: France
Posts: 692
|
OK, I though you said "video-sync=audio" was a reclock clone, and I just wanted to say it's just the default mode.
For "video-sync=display-resample", yes it changes video and audio like a Reclock lite indeed, but we can't say it's a reclock clone since it is very limited. As you said, it must be something like some milliseconds or it does not change anything, whereas Reclock was able to change 23,976 to 24 fps. For madVR smooth motion, maybe it's "laughable easy" but as you can see, developers did not understand it as you did, and there are big dev' names in this github ticket. So as you can see, there's no real madVR smooth motion clone. There's something which is close (resampling "Reclock Lite" with interpolation "oversample"). But in the end, there's still this "hiccups" problem for 60 fps on 120/240 Hz for every media players. Thankfully I have a script to change display refresh rate with a shortcut in order to have refresh rate = video fps. That's strange that you don't know mpv that much, with all your knowledge, I'm sure you would have fun testing all settings. It just requires some time reading the manual entirely. You could even report useful issues (like HDR stuff I don't know anything about for now), and I'm sure guys like Haasn would answer you, even with test builds to try. |
7th February 2024, 00:39 | #14 | Link | ||
Registered User
Join Date: Oct 2012
Posts: 7,925
|
Quote:
Quote:
you wouldn't believe how stupid i am and you wouldn't believe how little i know about HDR tone mapping. |
||
8th February 2024, 19:16 | #15 | Link |
Registered User
Join Date: Oct 2012
Posts: 7,925
|
yes that you where right mpv is lot of fun.
> > . you write a OETF into a EOTF field the gAMA and say it is fine > > Because the spec tells you to write 45455, not 2.2. If you think this is wrong you should take it up with the PNG specification. > > > because we also write the cICP > funny because 0.45455 is the reverse function of gamma 2.2 not 1.961. so the spec tells you to write 2.2 and you write 1.961. > Indeed, the PNG specification tells you that `cICP` overrides `gAMA`. > > > which is bt1886 which is 2.4. > > No, neither BT.709 nor BT.1886 are pure gamma curves, and can't be described by a single gamma value. > video gamma is a pure gamma function and can be defined by a single number and yes bt1886 can be to which is 2.4 inf CR. that's the beauty of it while the actual gamma user get's can be pretty much anything the rendered mathematical gamma is 2.4. and yes the EOTF in bt 709 was at one point 2.2 or 0.45455 > > or with other words: can't read the cICP > > Then you need to either find a PNG viewer that can (I hear chromium does! as does ffmpeg/mpv). Alternatively, you could use `--screenshot-tag-colorspace=no` which will coerce the image into sRGB before writing the PNG. yes just turn stuff into sRGB... which has a higher gamma then bt1886 according to this software only of cause. 2.2 is bigger then 1.961 last time i checked. |
8th February 2024, 19:37 | #16 | Link |
_
Join Date: May 2008
Location: France
Posts: 692
|
Ahah, I have no idea what you're talking about (remember, I don't understand anything between sRGB , bt1886, cICP, OETF, gamma) but I surmise that you had a raging discussion with someone. Was it on their IRC channel ?
As you can see in this ticket, they can have a disdainful tone. If you have questions on tone mapping (I don't know if it's what you're talking about), you can find informations here. I know it looks like it has nothing to do with mpv, but JRiver video renderer is based on libplacebo (if I'm not outdated), which mpv also uses, and the creator of libplacebo (haasn) was present on this thread, debating with some people, including a random guy called nevcairiel. |
8th February 2024, 20:19 | #17 | Link |
Registered User
Join Date: Oct 2012
Posts: 7,925
|
the tone is not the problem the shear idea of a gamma of 1.961 for video not oetf and just closing it and running away that i have a problem with.
https://github.com/FFmpeg/FFmpeg/bla...til/csp.c#L135 same person... if you find a person online that calibrated there display to gamma 1.961 (my meter wouldn't good enough for this BTW.) i will give you a nothing in return. i also have a feeling that random nev may have a proper port of super xbr for JRVR not like the one for mpv. just a feeling. |
8th February 2024, 21:06 | #18 | Link |
_
Join Date: May 2008
Location: France
Posts: 692
|
Guess what huhn, I'm still lost and I can't accuse your charming english !
But I remember you opened tickets on madVR tracker, it would be nice if you opened tickets on mpv tracker. I can't guarantee a positive ending, but at least there's a chance of improvment, because speaking with me or IRC channel has nearly 0 chance to achieve something. :/ |
8th February 2024, 22:00 | #19 | Link |
Registered User
Join Date: Oct 2012
Posts: 7,925
|
that is literally an issue tracker.
where i try to explain that bt709 is not gamma 1.961 and a more real gamma should be used. which by just writing makes me suffer. the "problem" is actually ffmpeg here. by not differentiating between oetf and eotf. i actually don't have a fix for ffmpeg but a video renderer should not use a OETF for a video file. |
9th February 2024, 00:37 | #20 | Link |
_
Join Date: May 2008
Location: France
Posts: 692
|
You did not pinpoint to the ticket number, if someone wanted to check the whole (unfruitful ?) discussion ! Maybe the dev is wrong, but maybe it's you ? I didn't say that you are wrong, but sometimes I am dead sure of something and it's hard to reason me.
If there's a bug in ffmpeg, then there's a bug in mpv, mpc-be, JRiver so I'm sure people would be interested in this discussion. Anyway don't give up on mpv, it's a good media player (mpc VR too) and it would certainly better if you reported issues. |
|
|