I dont know much about this and I'm stupid and paranoid about video when I can change these things myself What do these settings do? What is their aim? They try to achieve maximum fidelity to the original video, right? So, like, if you upscale it on a larger screen, its indistinguishable from watching it nonupscaled on a screen as big as the video, and everything is rendered perfectly faithfully otherwise too, yes?
Robert Perry
>They try to achieve maximum fidelity to the original video, right No. Things like sigmoidized scaling and debanding are deliberately ignoring fidelity for better subjective appearance.
Hunter Williams
what settings should I useif im autistic and want maximum fidelity like I described?
Austin Gutierrez
>MPDN - GPU Usage: ~71% - Chroma Upscaling: Bilateral - Luma Upscaling: Sinc-Blackman 12 Taps - Debanding: Default Strength / No Grains - Display Changer: Activated - ReClock: Activated - FluidMotion: Activated - Results: Smoother than with mpv with a little bit sharper picture but fans are very noisy.
>mpv - GPU Usage: ~92% (100% with some files) - Chroma Upscaling: spline36 - Luma Upscaling: spline36 - Debanding: yes - Display Resample: yes - Interpolation: yes / linear - Results: Less noisy fans than with MPDN but poor performance, get dropped frames regularly.
Do GPU fans act differently with directx and opengl?
Parker Hernandez
I guess things that go for their personal judgment about what would get a better subjective appearance are probably better for something badly encoded or having somehow lost quality(aside from being just lowres, in which case again, simulating what it'd look like on a screen it didnt need to upscale for with as high fidelity as possible would be better). But, still, I'd not prefer it for anything good to begin with. Autism, yes, I know.
Sebastian Bell
Even upscaling doesn't have a real answer to this, because video frames generally aren't really a collection of single point samples of a band-limited image. Talking about what's mathematically ideal doesn't have much relevance to reality.
It's easier to list thing that deliberately ignore fidelity: sigmoidization (makes lines look fatter) debanding (adds noise, even with deband-grain=0, blurs subtle details) interpolation, depending on what you think is the correct interpretation of frames (are they really samples in the sampling theorem sense? In the case of hand drawn animations in anime, certainly not) (blurs motion) antiringing (blurs some details) sharpening filters (adds ringing) NNEDI scaling (just blatantly making shit up)
But then again, is the video file you have really considered the reference video? Eg. commercial video is generally produced by people far less autistic than MPVthreadposters. For us it's personal, but for them it's only a job. Naturally they'll fuck it up. Maybe you should imagine what the original really looked like and tweak the settings to try to replicate that. The fidelity ignoring options can help with that.
Brayden Taylor
EH, I get what you're saying, and I feel fine with it for my cheaply animated, mediocre art animes, but I still get the obsessive pangs of paranoia and 'this cant be right' when I think about using it on actually well drawn, well animated, high video quality downloads.
Asher Kelly
>download mpv >get the autoload lua script from the github >mpv skips from episode 10 to episode 100 >delete mpv
Justin Adams
>not numbering your episodes with leading zeros
Matthew Hall
Do a double blind test in motion (and not on test images) and I doubt you'd be able to tell the difference with:
deband on default sigmoidization off interpolation off none of the meme shaders on
and the exact same setup with deband off.
The most obvious difference is interpolation on and off. If not having to deal with judder means things might get more blurry, so be it. Chroma scaling can cause issues on rare edge cases (not applicable for real life content).
To actually be true to the source you need knowledge that you cannot get. What was the lighting conditions that it was mastered in. What was the gamma curve used. What monitor was it mastered on, and how much deltaE were each of the colors off on even after calibration. Did it suffer from any downsampling loss when it was converted to an intermediary codec not in 4:4:4 like Prores 422. And what about the software. The gamma curve in a few professionally popular production software rounds in certain places.
It's best to simply add subjective enhancements that might ignore fidelity to approximate the source.
Benjamin Lewis
>be me >been using interpolation for months >suddenly I play my old videos >literally HEAR audio resample, like on the old audio cassettes
Well, goodbye interpolation
Cameron Flores
I have a i3-4130 cpu, r9 380 gpu and 4gb ram.
This is my setting that I'm playing with: vo=opengl-hq:backend=x11:scale=ewa_lanczossharp:cscale=ewa_lanczossharp:temporal-dither:interpolation:blend-s$ video-sync=display-resample #dscale=mitchell:tscale=oversample:scaler-resizes-only:dither-depth=auto #swapinterval=1:correct-downscaling:sigmoid-upscaling:pbo:deband #cache=262144 #hwdec=no #:icc-profile-auto:approx-gamma screen=0 autofit-larger=100% audio-display=no framedrop=no
The ones commented are those I'm not sure if I should add or if I'd need to set hwdec=0 so they work. What to do for good interpolation? (when I pause, in the terminal I get the following message :"Audio/Video desynchronisation detected! Possible reasons include too slow hardware, temporary CPU spikes, broken drivers, and broken files. Audio position will not match to the video (see A-V status field).
AV: 00:04:08 / 00:49:09 (8%) A-V: 0.000 "
Xavier Baker
Forgot to mention, I'm on Debian sid
Jaxson Miller
>lua script doesn't work the way I want it >ugh, I hate this media player! Good. mpv is not for mouthbreathers.
Parker Sanchez
Configure your OS for soft real-time performance. Look into cpusets and process isolation. Audible changes in audio most likely mean you have excessively high jitter, which means some problem with real-time performance.
Henry Moore
Post a clip on github or fuck off, fearmongering cunt.
Landon Roberts
Usually caused by broken files. Try remuxing to MKV.
Lucas Williams
>ffmpeg with a small change and a modified header. What are these changes? They're not documented as far as I can see
Ayden Phillips
Why does mpv need display-resample for interpolation where madvr and mpdn don't?
Carson Clark
Does anyone know if it's possible to load scripts in SMPlayer?
Jonathan Brooks
> This requires setting the --video-sync option to one of the display- modes, or it will be silently disabled. This was not required before mpv 0.14.0.
Logan Murphy
anyone on the macOS Sierra beta?
if so, could you please try this?
>add this to input.conf Shift+MOUSE_BTN3 seek 5 Shift+MOUSE_BTN4 seek -5
then, scroll up with the mouse wheel (not with a trackpad or magic mouse, an actual mouse with a mouse wheel) while holding Shift.
does it work fine?man, here it only seeks backward
Josiah Gray
Does mpv work with mesa? Does it perform like the official AMD drivers on windows?
Interpolation doesn't work on windows. Could I expect a better result on linux with mesa?
Kayden Hernandez
>Does mpv work with mesa? >Does it perform like the official AMD drivers on windows? Yes and slightly better since mesa is the reference implementation for gl on Linux.
The Sourceforge .7z with mpv includes an updater script written in Powershell. Those commands, when put into the command prompt, update mpv, though I'm not sure what %~dp0 means.
Easton Rodriguez
That's a problem with your filenames. Windows and linux both sort files alphabetically and if your files are not named 001, 010 and 100 they will load in wrong order in every player that doesn't parse filenames for numbers and somehow smartly sorts around them.
Hunter Myers
>The Sourceforge .7z with mpv includes an updater script written in Powershell. Those commands, when put into the command prompt, update mpv
Nice! This is exactly why sourceforge builds have BTFO the competition. Best Installer. Best Set-Up, Best Updater.
Landon Barnes
Only one is
Jace Richardson
>no longer works on xp
What the fuck did they do?
Isaac Bailey
>What do these settings do? What is their aim? Depends on the setting. It's too hard to answer this question in general.
I disagree. Sigmoidization significantly decreases ringing with no discernable drawbacks, which brings you much closer to the source. (Unless, of course, you enjoy your credits text being all messed up and think it's somehow “authentic” that way)
Debanding exists to counteract quantization, which is part of the encoding process. I would personally argue that turning debanding on is closer to the source than turning it off, simply beacuse the source got quantized while encoding and debanding more or less reverse that.
Plus, debanding doesn't really have any discernable drawbacks (at the default settings)
Thomas Torres
GPUs are too complicated to generalize like that. Can't answer your question.
As for your severe mpv performance issue, is this using ANGLE or dxinterop? (Also, can you try the performance on Linux? mpv performs pretty badly on Windows)
Top: Source (nearest neighbour) Middle: Sigmoidization on Bottom: Sigmoidization off
Ayden Ramirez
Well, really you have to define fidelity here. The problem with questions like these is that “fidelity” is not well understood for video.
In contrast, we understand audio fidelity fairly well (THD, frequency response, noise floor, crosstalk, etc. etc.) and there are objective “betters” in every category. You can easily compare, say, two resamplers and easily answer which one of the two performs “better” for any given metric.
Video, on the other hand, is not well understood; most promimently because we don't have good video quality metrics. There's stuff like PSNR, SSIM, etc. floating around but none of them really capture how the eye works very well.
Because of our lack of understanding for how to quantify visual quality, the best we can really do right now is simply be subjective in our metrics.
Julian Hall
Debanding on/off is very trivial to notice on banded test clips.
You can hear a difference of 0.1%? Holy shit, I think you have superhuman hearing
Asher Nguyen
I have poorer performances with dxinterop so I kept angle. I would like to try ubuntu or manjaro as it seems they are the easiest distro to install but it's a little bit retarded to use linux only for one software that has a similar alternative that works pretty well. Are there any chance to get better performance on windows in the future?
Austin Anderson
Okay, it seems that I can hear it in the moments, when I open the file, or in the moments of rewind. In next moments either I accustom to it, either it becames normal.
Jaxson Peterson
>mpv performs pretty badly on Windows ANGLE is actually faster than opengl.
Mason Jones
>There's stuff like PSNR, SSIM There is also new metric VMAF (by netflix)
Joseph Ward
Eh time to use Sigmoidization.
Landon Wilson
>I have poorer performances with dxinterop so I kept angle. Yeah, sounds like your drivers are pretty bad. ANGLE is usually the worse performing of the two (it translates all OpenGL calls to DirectX at runtime, which is pretty suboptimal).
>Are there any chance to get better performance on windows in the future? Maybe with vulkan, but I wouldn't get your hopes up. Microsoft hates open source software with a passion. I expect them to gimp Vulkan-on-Windows like they gimped OpenGL-on-Windows just so people stay trapped inside their walled garden infrastructure.
Brayden Hill
What mpv version and AO?
If your OpenGL drivers are completely shit, yes. AMD or nvidia?
Carson Edwards
nvidia ANGLE is aways faster
Brandon Jenkins
>Yeah, sounds like your drivers are pretty bad. ANGLE is usually the worse performing of the two (it translates all OpenGL calls to DirectX at runtime, which is pretty suboptimal). if you never tried it, don't make shit up.
Austin Butler
m8 I've been in these threads since ANGLE began to exist as thing and the universal response has been for people with non-broken drivers to get better performance out of dxinterop compared to ANGLE.
You're free to measure it yourself though, check with T show-text ${vo-peformance} during playback.
Liam Reyes
Do you mean it's faster than running the "native" opengl on windows or that's it even faster than running opengl on linux?
Isaac Brown
I have literally benchmarked it an hour ago. without shaders - same gpu usage with ANGLE has lower gpu usage
opengl performance on windows = opengl performance on linux for me.
Luke Flores
(typo) *with - ANGLE has lower gpu usage
Kevin Sanders
>What mpv version and AO? the latest build from here mpv.srsfckn.biz/ And I didn't quite understand, what's AO.
Chase Harris
Why is it that MPC-HC with Reclock and madvr smooth motion are much better than mpvs video-sync=display-resample interpolation tscale=oversample?
Might be a windows-specific audio timing bug. You could check with --dump-stats I guess
Grayson Davis
Try the builds from the link in OP they are much better.
Jose Miller
>56546743 Why is it that MPC-HC fags come in here and ask dumb questions? Not even getting a (You) from me.
Andrew Bell
MadVR doesn't really supports ReClock and the timings are completely wrong, you can't trust them. :\
Jaxon Nguyen
Is a "nvidia geforce gtx 960 2go" is enough for interpolation with new 4k videos?
Caleb King
>debanding doesn't really have any discernable drawbacks (at the default settings) Default banding is too strong, blurring details, and has noise enabled, which is also visible and pointless.
Michael Thomas
Can you show an example of where default debanding degraded the image quality? I somehow don't believe you. The defaults are extremely mild.
Owen Thomas
The sigmoid center default to 0.75 so it shows up better with dark on light. Here is the no sigmoidization version:
Carson Campbell
And here is default sigmoidization. I do think it looks better, but the lines have grown fatter. Flip between the images to see.
Aaron Evans
I still liked bitbucket builds better. It's in case you want to update it trough the CMD like I personally do
Caleb Cruz
How can I compile mpv with vapoursynth on Wangblows?
Evan Scott
RGB PNG input, full scale gradient, one pixel per color. Default scaling, default deband with grain disabled, 8 bit dithering. Nice and smooth. Dithering is not annoying.
Camden Murphy
Same thing with default deband-grain. Note how it has introduced obvious vertical axis noise that wasn't there originally. It has an ugly blotchy look but it's no smoother than the deband-grain=0 version.
Ryder Price
The default dither-grain setting of 48 equates to a difference of +/- 0.7 to the pixel value. 8-bit dithering is a difference of +/-0.5.
If you think dithering is fine but the default deband grain is too strong, you are imagining things.
Brayden Murphy
You are completely missing the point. Look at the spatial frequency of the noise! This is upscaled 3x size. Debanding is done pre-scaling.
Angel Scott
These two images have a mean absolute channel distortion of less than 0.001 (1 part in 1000). That's about 2^-10, which exceeds the human visual capacity except in extreme controlled circumstances.
Furthermore, those two images look completely identical to me on my 10-bit calibrated IPS panel even when magnified and squinting.
I see none of what you describe as “lines”, let alone any sort of visual grain. Have you considered that perhaps it's not the image, but your display?
Mason Cook
>I see none of what you describe as “lines” Where did I mention lines? Are you confusing this with the sigmoidization discussion?
And yes, the magnitude of the noise is similar, but the frequency distribution is not. And because the human visual system is less sensitive at high frequencies this makes the quiet version look noticeably better.
William Hernandez
Peak absolute error between the two is about 0.7% (PSNR of 50 dB), and happens in these places
Jacob Watson
Sorry, I misread and thought you were referring to noise that crossed the entire vertical axis (i.e. lines)
Robert Powell
If you really can't see it, and if you really think that's normal, that's still arguing for deband-grain=0 to be the default.
Either I'm right and it harms visual quality, or you're right and it adds GPU load for no perceptible difference.
Colton Wilson
I can't see any difference except that this image is 300% bigger in file size.
Gavin Perry
Arguing that deband grain is pointless because of one synthetic test pattern is sort of silly. It's supposed to be barely visible by default (that's why the value is so extremely low), the only purpose is to help mask quantization artifacts.
Since your synthetic test ramp was not quantized, don't expect any quality difference. Try it on a heavily banded source file. (Or wait a bit, I'll upload some examples)
Juan Johnson
>blurring details I retract this, default strength seems to be conservative enough that blurring isn't a problem. But the default grain is definitely a problem.
Jacob Hernandez
What's opengl-pbo ?
Carson Sanchez
Sorry, I didn't make it clear that "obvious" was the important word. There's just as much noise in the horizontal axis, it's just hard to see because of the horizontal gradient.
Camden Reed
I'm on Debian 8.5 stable. mpv crashes really often when I want to go full screen always when I press F
William Gonzalez
use it if you are on nvidia gpu, and don't use angle.
Ayden Wright
>Or wait a bit, I'll upload some examples Bleh, I made some samples but they're too large to really upload; and if I use smaller cut-outs the banding isn't really visible (because stock deband gets rid of so much already), plus the effect is really hard to see in an image to begin with because the grain is dynamic while the banding is not.
Jaxson Clark
Also, I turned down my lights in my room and looked at your samples again. I can definitely see the grain now, but only in the very very dark region (i.e. towards the left strip)
In general, the darker the signal, the more obvious the grain is. Can you confirm this? Maybe I should gamma-adjust the grain to make it constant-luminance instead of constant-signal.
It's probably easier to see the higher your display's contrast is, which might explain why I'm having difficulty seeing it on my IPS panel.
Adrian Gray
wtf happened to the bitbucket repository?
Jordan Carter
haasn is one
Gabriel Rogers
Yes, I agree. Also when I'm watching videos I frequently increase gamma just to make things brighter, which would make it even more obvious assuming deband is done before gamma correction.
Connor Howard
Deband is done on the input, in the file's native color space (i.e. the same one that it was quantized with). The idea is to mask quantization artifacts on a source level, before upscaling smooths it out.
Btw, have you tried using e.g. ewa_lanczossoft instead of the default opengl-hq scaler (spline36)?
Kayden Green
I use one of the 3 ewa_lanczos variants (smooth/default/sharp) depending on how aliased the source is.
Jordan Anderson
How do I exit full screen? srsly I tried everything!
Only thing that worked was ctrl+esc and end process, but then taskbar on top of the windows in kde disappeared
Noah Perry
>srsly I tried everything! clearly you haven't tried pressing the ‘f’ key
Kevin Gomez
why esc wouldn't exit full screen? it makes perfect sense. Fucking Linux dauns
Ethan Ortiz
esc works too for me. You are probably just a daun.