Search options (Page 8 of 349)

вон там - https://www.svp-team.com/forum/viewtopi … 672#p84672 - .dll которая, возможно, это исправляет
пока только для Vapoursynth

https://www.svp-team.com/forum/viewtopi … 672#p84672
updated svpflow2_vs.dll, default value for scene.limits.blocks set to 40 for RIFE

> Then how to make proper frame blending when RIFE_SC threshold over 15?

this is why you need a MV-based scene change detection instead of dumb image comparison

> NVOF algo fixes vertical motion artefacts but causes problems elsewhere.

It just can't "cause problems". It can only:
a. (_very_ unlikely!) miss the scene change that is actually a scene change --> which shows RIFE's artifacts at this frame
b. mark a scene change where there's no scene change --> makes it choppy where it could be not choppy

that Hugo?

Testing RIFE update with SVP's scene change detection added.
Replace script\generate.js, script\base.py and plugins64\svpflow2_vs.dll, restart SVP.

Now THREE different modes are possible, controlled by the "Scene change threshold" profile option:
1. default - old behavior with "dumb" scene change
2. scene change based on SVP's motion vectors - when "scene change threshold" is "6%"
3. scene change based on NVOF's motion vectors - when "scene change threshold" is "8%"

Scene change-related params from override.js should also work, such as smooth.scene.limits.zero, .blocks, .scene (see a short description there - https://www.svp-team.com/wiki/Manual:SVPflow)

UPDATE: svpflow2_vs.dll updated, also supporting blending on scene changes instead of frame repeating.
There's no "processing of scene changes" option in the RIFE profile so you have to set it via 'All settings' - find the profile and set 'fi_scene_changes' to 0 (blend) or 1 (repeat)

UPDATE 2: svpflow2_vs.dll updated, default value for scene.limits.blocks set to 40 for RIFE

UPDATE 3: svpflow2_vs.dll updated, using correct frame for repeating on a scene change

> I can't actually find any ncnn vs dml benchmarks for rife.

TRT is 2 times faster than ncnn
DML is _10_ times slower than TRT

flowreen91

step 6: "enjoy" what exactly? 10-15% performance drop?

another post that is actually a bad advice for people who don't understand what their doing, just following "guides"

----
dawkinscm
> If you have to use Rife v2 and have picture shaking issues then updating vslmrt makes sense.

only for TRT>=9
with TRT8 updating vsmlrt.py most likely does nothing
only difference between SVP's vsmlrt.py and current master is one line:

# TensorRT 9.0 or later
"ONNXTRT_Broadcast_*:fp32"

one more time, what is the exact point of "v2"?

it's not-faster or even slower, and it takes more RAM. so, why?
just because "v2" is cooler than "v1"? also tensort rt v10 is cooler than v8...

re(install) Avisynth+ manually
https://github.com/AviSynth/AviSynthPlu … redist.exe

поискать в драйверах Интел "улучшатели" видео, и всё выключить

> Huge improvement over ncnn for Radeon and Arc.

any benchmarks?