I'd suggest updating to the latest version first...

avisynth filter + madVR

it works since "4.3.0.191 – 2020-04-11" because SVP enumerates all mpv processes and tries /tmp/mpvSockets/<pid>

just take this - https://github.com/wis/mpvSockets - and that's it

ну вон DAR правильно выставляется в 1.3333
что, если потом в плеере открыть, то неправильно показывает?

вот для примера:

14:48:28.348 []: Media: video 720x464 [PAR 1.422] at 25.000 fps
14:48:28.348 []: Media: codec type is AVC, YUV/4:2:0/8 bits
14:48:28.355 []: Transcoder: Frame server (64-bit) C:\Program Files (x86)\SVP 4\mpv64\vapoursynth.dll
14:48:28.355 []: Transcoder: resulting video frame 720x464
....
а в логе кодировщика:

14:48:33.669: Command line: C:\Program Files (x86)\SVP 4\mpv64\mpv.exe D:/share/test video/Дни Затмения [1988]_sample.mkv --o=D:/tmp/Дни Затмения [1988]_sample.SVP.temporary.mkv --no-audio --no-sub --no-sub-auto --input-ipc-server=mpvencodepipe --input-media-keys=no --no-msg-color --vf=vapoursynth:[C:\Users\chain\AppData\Roaming\SVP4\scripts\ffff.py]:4:16,format=dar=2.20474 --of=matroska --ovc=h264_nvenc --ovcopts=b=3223155,preset=slow,profile=high,rc=vbr,maxrate=4190102,bufsize=6446310,time_base=1000/71875,threads=16

720 * 1.422 / 464 = 2.205

логи надо. перекодировщик должен сохранять aspect ratio при не-квадратном пикселе, это специально исправлялось и проверялось.

1,533

(3 replies, posted in Эксплуатация SVP)

из 25 совершенно точно получится не 59.94, а 60
сделать можно что угодно, но делать именно 59.94 из 25 - плохая идея

1,534

(3 replies, posted in Эксплуатация SVP)

по умолчанию если ничего не трогать то будет 23.976 * 2.5 = 59.94
и только если поставить галочку на "точное значение", то будет 60

1,535

(9 replies, posted in Using SVP)

dunno hmm
the "Avisynth Filter" installer script sets all the registry values on installation/update
and I just checked - it really does

1,536

(1 replies, posted in Using SVP)

do you have /usr/local/lib/libvapoursynth-script.dylib ?

---
in fact it must be here: /usr/local/opt/vapoursynth/lib/libvapoursynth-script.0.dylib
and it will be here if you install Homebrew in a default location

> Maybe the moves to Nvof 2.0

"nvof 2.0" was added in 445.87

---
I have 460.79 on my laptop with gtx 2060, never seen this error too

---
dunno, I can't reproduce this with 460.89 neither on a 2060/laptop nor 2070/desktop

поправить "все параметры" -> cast.encode.opts.h264_nvenc = "b=$B,rc=cbr,preset=slow,profile=high"
"cbr_hq" поменять на просто "cbr"

1,539

(9 replies, posted in Using SVP)

most likely you don't have python 3.8 installed
libsvptube.so depends on python 3.8

1,540

(9 replies, posted in Using SVP)

os? any errors in the log?

1,541

(4 replies, posted in Using SVP)

halo is the most common artifact for any motion interpolation, you can't get rid of it with "increase bitrate and change encoders"
you can only play with the manual SVP interpolation options to get a result that looks good enough for you
there're numerous topics on this problem within this forum - use search

not sure it's related to the "Avisynth filter quality"...

btw, this error code means nothing - i.e. "indicates that an unknown internal error has occurred" while creating GPU memory buffer
to be clear: it comes not from our (SVP's) code but from the video driver

1,543

(4 replies, posted in Using SVP)

> are those sliders only valid during realtime playing?

yep, it always selects the highest quality in transcoding mode

> Do these improve the quality of Optical Flow over Turing GPU's?

no idea... if yes then it's already "improved" with the drivers update

try Avisynth Filter instead of ffdshow, it works much better when seeking
https://www.svp-team.com/wiki/Avisynth_Filter_(AVSF)

SVP doesn't support 1x1 motion vectors right now
and I really don't think it should
from what I read that 1x1 vectors field is just the same hardware-interpolated 4x4 field

1,547

(3 replies, posted in Using SVP)

the thing is if the source video has a bitrate X and you re-encode it with the same bitrate X - you won't get the same visual quality as the source

I think we're talking about _performance_. Blackfyre has a performance issues playing 4K @4px blocks, and this is NOT because his 2070 is too slow in NVOF.

And I don't think 1x1 blocks are useful for us. 4x4 is more than enough.

> Anyway, so how can I achieve 4px with 4K HDR content?

I think it's more CPU-related

> Or it doesn't matter which RTX videocard because they all perform the same with Optical Flow?

yep

mmm... nvof_grid = 0 gives a script error with "division by zero" and nvof_grid = 1 gives another script error

in fact, NVOF engine works with 4*4 px block only wink
so "4px" in SVP is just "search 4*4 blocks on the original frame size"
and "8px" --> "search 4*4 blocks on the frame reduced by 2 in both directions"