New RIFE filter - 3x faster AI interpolation possible in SVP!!! (Page 61 of 62)

Each iteration after 4.4 causes a massive performance hit @~16% for minimal artefact improvement. Above 4.9 is trained mostly on Anime.

What you say doesn't make sense to me. You say you don't use 4.4 because "too much artefacts (not critical but noticeable)", but then you are happy to force downscale all 4K and 1080p content to 1280x720p?! And you don't notice the huge degradation in quality?

I am guessing you don't have a 4K display, or else you wouldn't be saying something so seemingly bizarre

To each their own, I guess...

Who said my monitor is 4K? I dont watch 4K at all, my monitor is standard chinese Full HD 165Hz GSync with builtin upscalers, I just test 4K source for performance, hehe! I dont see massive quality degradation only because of monitor upscalers, but my solutions shown are exclusively for 6x movie fps targets, obviously I watch 1080p with 3x for sure!

There was a bug with Rife v2 which has been fixed since vslmrt script 3.18.4 released November last year. I think this bug caused picture shake issue.

I know you are not replying to me but I think your comments are interesting so I hope you don't mind me responding directly to you. As you say, there is a massive performance hit. But you have been very lucky if you haven't come across the horrible artefacts that appear below 4.9.

Being "trained on Anime" or not doesn't matter if it still results in improved live action motion. Rife v4.9 and above are all smoother and have much less visible artefacts than below 4.9.

Resolution is important but only up to a point. Although I personally would not downscale to 720p I do now downscale my 1920x2160 3D blu-rays to 1080p.   On a near IMAX VR screen, I've tried both upscaling to 4K and downscaling to 1080p and found no "visible" difference between the two. For actual 4K content there is a difference, but it will only be a problem on very large of screens. For most people watching on a TV, there will be little visible difference unless you decide to stand in directly in front of the screen.

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...

While it's no longer significantly faster, the release notes say it reduces "PCie traffic flow". From personal experience this helps a lot for real time processing and provides more GPU headroom. Before I started to use downscaling, I was regularly hitting 100% GPU, but the v2 models would bring that down to 90% or less depending on the model.

But @Chainik thanks for the reminder because it's been a while since I used a non v2 model and unlike previous models, v4.15 seems to handle artefacts a "touch" better than the v2 version.

How can I change the TRT version? And how can I find out the current version? I really don’t like the scene change on any model at any 6%-100% values. If the dynamic scene changes to dynamic, then some kind of jerk occurs, it is instantaneous, but it is visible. Maybe they fixed this in other versions of TRT?

> How can I change the TRT version? And how can I find out the current version? I really don’t like the scene change on any model at any 6%-100% values.

this isn't related to the TRT version at all
TRT version can only affect build time, memory usage and overall performance

Guide for updating RIFE to latest TensorRT from 3 days ago: https://github.com/AmusementClub/vs-mlrt/releases

Step 1: Download https://github.com/AmusementClub/vs-mlr … uda.v14.7z
Step 2: Go to C:\Program Files (x86)\SVP 4\rife\ and move the following to a backup folder: vsmlrt-cuda,vstrt.dll,vsmlrt.py
Step 3: Open vsmlrt-windows-x64-cuda.v14.7z and copy vsmlrt-cuda,vstrt.dll,vsmlrt.py in their place

Step 4: Edit vsmlrt.py replace line 1821
old:
tempfile.gettempdir(),
new:
os.path.expandvars("%APPDATA%\\SVP4\\cache\\"), #!SVP

Step 5: Edit helpers.py replace line 52
old:
return RIFE(clip,multi,1.0,None,None,None,model_num,backend,ensemble,implementation)
new:
return RIFE(clip,multi,1.0,None,None,None,model_num,backend,ensemble,True,implementation)

Step 6: Enjoy 10-15% performance drop caused by upgrading to latest code!

Noteable issues:
Your PC might not have the power to 2x a 4K video, and mpv might display desync errors that cancel SVP interpolation https://gyazo.com/a4714c85227b5b596a5f71084b5c2fae
so u should test it with a frc.frame.resize -19201080 or lower

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"

Noted thanks. I'm currently using v4.15 but if I go back to a v2 model then I will update TRT. Cheers

You're just going to have to experiment and see what works best for you. Most of my movies work well with almost any SC or even turning it off. But I've got a couple that need SC to be set to I had to find a balance.