976

(13 replies, posted in Using SVP)

none is better or smoother
difference is more threads could rise GPU utilization in case it's too low while output fps is not enough

а если сейчас?

ещё раз обновление запустить

> Although, anywhere above 48 fps my GPU usage sky rockets from 46% to 90%

because x2.5 interpolation (i.e. 24->60 fps) need generating TWO times more intermediate frames than x2 (24->48)

> I added another model from github and when selected I get a flickering sensation on any video

then add another model big_smile and post a bug report on the Github regarding the flickering one

980

(209 replies, posted in Эксплуатация SVP)

"Number of devices: 0" как бы намекает, что что-то тут не то

SVP updated wink

---
RIFE/CUDA removed from the script completely

> and leave 4.4

why 4.4? just because the numbers are higher than 4.0?

ok, 4.0 was removed from the list, but 4.2, 4.3 and 4.4 are still there, w/o any changelog or something like that
is 4.4 better than 4.2?

it won't work on a 7 yo SoC

> Do you plan to update SVP with the newer models?

kind of... not exactly like this
there will be only ONE model bundled (*), plus you'll be able to add whatever model you want manually, by copying model's folder

(*) if you _know_ which one from 4.0 to 4.4 is the best in terms of overall quality - this is the right moment to tell

> without losing 10-bit colour depth and HDR?

yep

> I would very much encourage everyone to share their observations

buy a video card at last! big_smile

> Is there an improvement in quality and speed of interpolation?

no idea about quality, but 4.1-4.4 performance is the same as 4.0
4.5-4.6 are 10% _slower_

987

(484 replies, posted in Using SVP)

-no-tray ? big_smile

988

(209 replies, posted in Эксплуатация SVP)

> на AMD Radeon SVP будет работать

если попрыгать с дровами, то будет

too bad... dunno what we could do about it right now
need to figure out some workaround hmm

actually I need what is below this - the list of CDM's

991

(4 replies, posted in Using SVP)

> But inside it's still 4.5.0.212

yes, but with the latest yt-dlp inside

992

(4 replies, posted in Using SVP)

there actually WAS an update a few weeks ago

---
you will see it now big_smile

please run the app called "DRM Info" (there's one in the AppGallery) and post results here (via screenshots, I suppose)

> SVP allows real-time interpolation for 4K 10-bit HDR files, but needs Avisynth Filter for this

"Avisynth Filter" is a confusing naming cause there're _two_ different filters there, one for Avisynth and one for Vapoursynth
both supporting 10-bit and hdr

> It is said that the performance of Optical Flow is also improved by about 2.5 times

it only makes difference for interpolating 4k@60 to a higher rates

> will AviSynthPlus-RIFE work with 4K 10-bit HDR...?

exactly the same as it work in vapoursynth-rife

>  would you be able to implement this new AviSynthPlus-RIFE into SVP

yeah, will do... but there's no noticeable difference with a "Vapoursynth filter" - load times and performance are exactly the same
plus it requires an unreleased AVS+ build, crashes with the latest official release

Just built vapoursynth-rife-ncnn-vulkan on Apple M1, and it only makes 50 fps out of 704*528@30 source video with v4 model.
Not exactly the performance I expected...
But at least it works, which is good news smile

да даже 4*4 иногда странно выглядит, еще мельче кмк особого смысла нет
плюс дальше обработка идет через CPU (пусть потом и заворачивается обратно на GPU), т.ч. в текущей архитектуре SVP вектора 1*1 просто положат проц, а на 4к так и 2*2

да и еще хз откуда они там берутся эти 2*2 и 1*1
в SDK для 2ххх они тоже были, но тупо считались интерполяцией из 4*4 в самом SDK
может в 3ххх это такая же интерполяция, но "спрятанная" в драйвер, мы ж не знаем big_smile
а если так, то смысла в этом никакого нет, SVP и так делает интерполяцию векторов до 1*1, но на самом последнем этапе

> Собственное удвоение кадров. Очень жду имплементации.

в минимальном виде это будет просто передвижение точек по тем же векторам, ну плюс какой-то интерфейс чтоб это где-то использовать, с ffmpeg например
но если выложат какой полезный алгоритм маскирования артефактов - тогда хорошо big_smile

куртка - читер big_smile