/mpv/ - ngu like algorithms

Is the mpv dev team working on a "next generation" scaling algorithm?

Other urls found in this thread:

github.com/bacondither/Adaptive-sharpen/commits/master
twitter.com/SFWRedditGifs

For me NNEDI3 32+ is currently better than madshi's NGU and I don't think haasn is going to create a NGU like algorithm. It's much safer with crappy sources. Haasn is not active anymore on github, hope he is working on something new for mpv.

KrigBilateral is really good but my gpu is in extreme pain.

Aadaptive sharpen was updated, don't know if it improves something:
github.com/bacondither/Adaptive-sharpen/commits/master

Hey igv, did you try to create cnn algorithms like ngu? Is it that hard?

mpv devs don't give a fuck about prescalers. They provided a user scripts framework if anyone wants to make their own doublers and whatever shit.

Why exactly? Do they think it's a waste of time or resource?

Could you or anybody else show me a comparison between NGU and haasnsoft. I am sure haasnsoft is better with less ugly artefacts.

Should I enable cache?

They have different reasons. wm4 simply doesn't care, haasn remarked before that "neural network" based and other prescalers distort original source too much into an "oil painting" like effect. There maybe some non-core mpv contributors like bjin and igv are willing to port them to mpv, if the algorithm is open source. They are not likely to invent or reverse engineer algorithms though

Only for streams

...

>wm4 simply doesn't care
Any plex premium users here to ask him? :)

>haasn remarked before that "neural network" based and other prescalers distort original source too much into an "oil painting" like effect
What about a better luma upscaler?

>they are not likely to invent or reverse engineer algorithms though
Why not?

Those algorithms are usually invented by academics with a background in signal and image processing and reverse engineering is dubious from a copyright standpoint. I don't know what sort of algorithm madshi has implemented, whether its novel or based on existing algorithms with additions so I can't comment on that

Madshi claimed NGU is not a simple implementation but how to check if it's true without the source code?

Fuck the copyreich and all these retarded scientists.

>ngu released
>faster than ewa_lanzossharp
>still not using madvr
>still complaining about oil painting
WHY?

just bought a 1060
what settings do I use to upscale all my shit to max tier?

currently using

profile-desc="High quality VO settings"
vo=opengl-hq:scale=ewa_lanczossharp:cscale=ewa_lanczossoft:dscale=mitchell:tscale=triangle:scale-antiring=0.8:cscale-antiring=0.9:dither-depth=auto:target-prim=bt.709:correct-downscaling=yes

can't see a difference though

What version of mpv are you using? Thats old syntax.

should be the latest version from the website itself

I copy n paste the config file from someone's github quite some time ago

this is what it should look like
profile=opengl-hq
scale=ewa_lanczossharp
cscale=ewa_lanczossoft
dscale=mitchell
tscale=triangle
scale-antiring=0.8
cscale-antiring=0.9
dither-depth=auto
target-prim=bt.709
correct-downscaling=yes

>Madshi claimed NGU is not a simple implementation but how to check if it's true without the source code?
Is it content neutral? How does it perform on a large variety of different sorts of content and what type of upscaling characterists/artifacts does it have? Compare it with known algorithms. Most of the cutting edge upscaling techniques used by new 4k TVs rely heavily on postprocessing with varying grades of sharpening, edge enhancement, noise/grain adjustments etc. Comments and comparisons on doom9 hint at madshi relying on a combination postprocessing techniques

Would it be possible to make some kind of css or extension for a browser to use mpv on youtube or other websites?

mpv + youtube-dl + openwith

If it is in a phd, it is in the public domain and should be available from the university's open access program for free.

I can never get yoube-dl working, can you give me a rough guide m9?

So I've had an issue that's just appeared in the past week or so. Basically if I open 'too many' mpv instances then Xorg starts freezing and it can take me up to 10 minutes to open htop and kill them all off so it recovers. The mpv instance that finally tips the boat always starts up invisible without rendering anything, however after that it can 'bleed' its problem onto the others (or not, it depends). I don't know if it's from a recent mpv update, a Xorg one, or the nvidia drivers but it's fucking annoying.

Put youtube-dl.exe in the same folder as mpv.exe

install windows 10

Is angle fixed now?

Haasn is gone, RIP mpv.

sores

His github status is quite clear, he is on vacation.
>I like Haskell and snow, they distract me from the living hell that is technology

Posting my mpv config.

# Video
correct-downscaling
cscale=ewa_lanczossharp
opengl-fbo-format=rgba32f
opengl-swapinterval=0
scale=ewa_lanczossharp
scaler-lut-size=10
sigmoid-upscaling
video-sync=display-resample
vo=opengl

# Audio
alang=ja,jpn,Japanese

# Subs
demuxer-mkv-subtitle-preroll
slang=en,eng,English

# Misc
autofit-larger=80%x80%
deinterlace=no
no-osc
osd-fractions
screenshot-format=png
screenshot-high-bit-depth=no
screenshot-template=~/shot%n
ytdl-format=[protocol!=http_dash_segments]

[altcolor]
target-prim=bt.2020
target-trc=srgb

[dvd]
hr-seek=yes
deinterlace=yes

[extension.VOB]
hr-seek=yes
deinterlace=yes

[image]
force-window=yes
hr-seek=yes
image-display-duration=inf
input-conf=~/.config/mpv/input-image.conf
loop=inf
loop-file=inf
mute=yes
no-keepaspect-window
reset-on-next-file=mute,video-pan-x,video-pan-y,video-zoom
script=~/.config/mpv/autoload.lua

just go there and see all the comparison screenshots

Mpv is officially killed by NGU. Rip.

But it's a dumb shit! Isn't it?

NGU? Another placebo meme scaler?