You are not logged in. Please login or register.
SmoothVideo Project → Using SVP → New RIFE filter - 3x faster AI interpolation possible in SVP!!!
different scene change detection methods work differently
I'm sure 
How would those differences be described, to allow an end-user like myself to know what might be reasons to pick one over the other?
Chainik wrote:different scene change detection methods work differently
I'm sure
How would those differences be described, to allow an end-user like myself to know what might be reasons to pick one over the other?
NVOF : good (smoother than SVP)
SVP : normal
RIFE : good/normal (weird changes depends on threshold but perfect with newer models)
> Is there anything we can do to increase NVOF and SVP algorithms strength
yep
smooth.scene.limits.scene
smooth.scene.limits.blocksthe bad thing is that values that work good with resizing to 720p doesn't seems to work with 1080p and 4K
Could you give more detailed info on how to increase the algorithm's strength when transcoding 1080p video, with examples?
Would it be possible to add an option in the GUI to simplify this process? I still find RIFE unusable with the hideous scene transitions.
NVOF : good (smoother than SVP)
SVP : normal
RIFE : good/normal (weird changes depends on threshold but perfect with newer models)
I think we're talking cross purposes here, the image comparison list in SVP has the following options
* Disabled
* SVP motion vectors
* NVOF motion vectors
* Image comparison
It's the differences between these four options that I'm looking for clarification on. Can you help?
RickyAstle98 wrote:NVOF : good (smoother than SVP)
SVP : normal
RIFE : good/normal (weird changes depends on threshold but perfect with newer models)I think we're talking cross purposes here, the image comparison list in SVP has the following options
* Disabled
* SVP motion vectors
* NVOF motion vectors
* Image comparisonIt's the differences between these four options that I'm looking for clarification on. Can you help?
Disabled - RIFE image comparison method (100% threshold) (same problems as image comparison with threshold >25%)
SVP motion vectors: smooth transitions between scenes but suffers with high speed camera movement (high frame repeat/frame blending)
NVOF motion vectors: smoother transitions between scenes than SVP but has wavy artefacts between scenes (low frame repeat/frame blending)
Image comparison: RIFE has weird scene transitions, overall smoothness better than others (high frame repeat/frame blending with threshold <5% / scene weirdness differs from model and threshold)
Rife v4.25 lite is now available for anyone who is interested.
Rife v2.5 lite is now available for anyone who is interested.
aka 4.25 lite
dawkinscm wrote:Rife v2.5 lite is now available for anyone who is interested.
aka 4.25 lite
Thanks
Rife v4.25 lite is now available for anyone who is interested.
Lots of artefacts in some test scenes. Similar to, or worse than 4.26
Back to my preferred ones below:
4.25 v2 < 1699
4.16 Lite v2 > 1700
RTX 4070 (10 months old)
4.25 V1/V2 Lite test > worst (the pixels interpolating separately from main object) (not happens from 4.0 to 4.18)
Test parameters: mod8/mod32/mod64 inputs | builder optimization levels 2/3/4
4.25_lite v2: max 5% less GPU usage than non-lite (4K HDR source material, frc.frame.resize=-19201080, fixed 60fps/SDR tone-mapped, upscaled to 4K with mpv/a couple of shaders) , no artifacts in my test scenes in The Martian, Jason Bourne, Rogue One, Big Bang Theory (1080p source, fixed 48fps) and downloaded YT-Video with different aspect ratio (fixed 48fps).
Smooth overall visual impression an 27" 4k monitor. Much better than 4.26x (which has many artifacts in my test cases).
Shaders:
hdebeband.glsl
SSimSuperRes.glsl
FSRCCNNX_x2_8-0-4-1.glsl
contrast.10.hk
Edit: I cannot post for another hour, wtf...
I think your shaders somehow interfere RIFE algorithms, because 4.26 was better than 4.25L during my test sessions, a lot, but pixel crosstalk still an issue!
What do you mean upscaled back to 4K thats output resolution now? (resize and upscaling interfere RIFE methods - tested)
Just to be sure, I just repeated the test again, the artifacts I'm testing/looking for are significant with and without shaders at 4.26 (as they are at 4.15l, 4.16l, 4.18l) and not visible at 4.25l (and 4.22l).
Yes, I meant the output resolution. Without downscaling for rife (and upscaling to output resolution of course) I wouldn't be able to get 48 or 60fps at all (GPU = 4070). I could downscale to slightly higher (e.g. 23041260, then max 48fps possible), but it's not worth it for my viewing pleasure.
While testing, I just noticed that if I set resize to -23041260, I see the artifacts at 4.25l and at 4.26. If I set it back to -19201080, only at 4.26. Erm... and -25601440 same as -19201080...
4.25_lite v2: max 5% less GPU usage than non-lite (4K HDR source material, frc.frame.resize=-19201080, fixed 60fps/SDR tone-mapped, upscaled to 4K with mpv/a couple of shaders) , no artifacts in my test scenes in The Martian, Jason Bourne, Rogue One, Big Bang Theory (1080p source, fixed 48fps) and downloaded YT-Video with different aspect ratio (fixed 48fps).
Smooth overall visual impression an 27" 4k monitor. Much better than 4.26x (which has many artifacts in my test cases).
Shaders:
hdebeband.glsl
SSimSuperRes.glsl
FSRCCNNX_x2_8-0-4-1.glsl
contrast.10.hk
I think your shaders somehow interfere RIFE algorithms, because 4.26 was better than 4.25L during my test sessions, a lot, but pixel crosstalk still an issue!
What do you mean upscaled back to 4K thats output resolution now? (resize and upscaling interfere RIFE methods - tested)
I think your shaders somehow interfere RIFE algorithms, because 4.26 was better than 4.25L during my test sessions, a lot, but pixel crosstalk still an issue!
What do you mean upscaled back to 4K thats output resolution now? (resize and upscaling interfere RIFE methods - tested)
Just to be sure, I just repeated the test again, the artifacts I'm testing/looking for are significant with and without shaders at 4.26 (as they are at 4.15l, 4.16l, 4.18l) and not visible at 4.25l (and 4.22l).
Yes, I meant the output resolution. Without downscaling for rife (and upscaling to output resolution of course) I wouldn't be able to get 48 or 60fps at all (GPU = 4070). I could downscale to slightly higher (e.g. 23041260, then max 48fps possible), but it's not worth it for my viewing pleasure.
While testing, I just noticed that if I set resize to -23041260, I see the artifacts at 4.25l and at 4.26. If I set it back to -19201080, only at 4.26. Erm... and -25601440 same as -19201080...
Hi everyone,
Given that the RIFE algorithm way of dealing with scene change is to duplicate the last frame, there are often instances in my testing, where the penultimate frame is very similar to the final frame and therefore looks like the video is frozen in time for a frame. Is there any way in which a different approach could be taken, that on the final frame of a scene the model ONLY looks at previous frames and not forward? So the final frame would be built from the previous?
Mardon85 wrote:Nope no dice. It looks like its dropping frames doesn't report that it is. I've just tried an AI Enhanced version of Wolverine Deadpool and I can notice the smooth motion immediately. At the same settings reported at 60FPS (Rife) it doesn't look smooth. If I turn on the OG frame interpolation, again I notice it (and the artifacts straight away).
I'm coming close to doing a clean windows install at this point.
I wonder if that enhanced version is my doing? ;D DirtyHippie
Sorry you are having issues with RIFE not interpolating as it should.
If I set MPC-HC to exclusive fullscreen and lock it to 60hz, it seems to play more smoothly from my PC. I always transcode the movies to 60fps, then play them from my android tv using KODI, as for some reason it plays better than from my PC. Another reason I prefer playing the transcoded movies from my TV is I can use true Dolby Vison vs the tone mapping from MPC Video Player in MPC-HC via Windows.
I don't think your issue is due to not having exclusive fullscreen set. Just though I'd share what I have discovered about that.
I am having trouble retaining Dolby Vision Metadata and the colour vibrance seems to change even using the settings you have posted. Could you help me with your Dolby Vision workflow? Have sent you a PM.
rife_v4.25_heavy - 153 MB
^^ Has been released around 5 hours ago. Tried to test it, but this is the first time I come across this problem; even when selected in SVP it does not work, it goes back to the previous version I had selected when running a video, so cache for "heavy" does not even build.
Not sure if something needs to be updated for it to work properly.
I only tested v2 btw, did not test v1
rife_v4.25_heavy works fine for me, both v1 and v2 versions, just like the other rife models - copy to folders, restart SVP and there you go.
I wasn't never able to run any non-RIFE models, but that's another story.
rife_v4.25_heavy - 153 MB
Not sure if something needs to be updated for it to work properly.
Try these for heavy?
Blackfyre wrote:rife_v4.25_heavy - 153 MB
Not sure if something needs to be updated for it to work properly.Try these for heavy?
Heavy ran first time for me with no issues. It didn't even ask to redo the engine model so I will probably need to make it do that. Removing cached entries doesn't help so I will try your files.
Update
Yes they work thanks. I don't see any difference with v25 in terms of artefacts. I haven't spent enough time to check out smoothness.
For me 4.25 heavy FIX smoothness issue from 4.11 to 4.25 models, my RTX 4070 runs these models choppy, not performance issue, no monitor issue, 4.25h now smooth as 4.10 without any artefacts!
What about performance? Well, 4.25 v2 heavy enough for 576p 24>192 conversion without drops (2 threads)
Try these for heavy?
Interesting - this vsmlrt.py breaks rife_v4.25_heavy v2 for me.
Guess I'll stick with SVP original.
rife_v4.25_heavy - 153 MB
^^ Has been released around 5 hours ago. Tried to test it, but this is the first time I come across this problem; even when selected in SVP it does not work, it goes back to the previous version I had selected when running a video, so cache for "heavy" does not even build.
Not sure if something needs to be updated for it to work properly.
I only tested v2 btw, did not test v1
Works VERY good here, it may be my new default... I don't get the naming convention, thus far lites have always performed better, but the heavy one is better quality and same or better performance than the lite one.
With my test cases of course.
I checked for any improvements in artefact handling but I couldn't find any. I already have smooth Rife, but it does seem to be a little smoother and performs well in my forward/reverse direction tests.
rife_v4.25_heavy - 153 MB
^^ Has been released around 5 hours ago. Tried to test it, but this is the first time I come across this problem; even when selected in SVP it does not work, it goes back to the previous version I had selected when running a video, so cache for "heavy" does not even build.
Not sure if something needs to be updated for it to work properly.
I only tested v2 btw, did not test v1
Same here, for me it use the 4.25 profile that was already build and not run another pass
Quick fix change the name of the model like 4.30 build engine run and create a new, testing...
For me 4.25 heavy FIX smoothness
Yeah I agree! this model is more smooth at least i gives me that impression.
BTW All this around the 4.25 model make felt good to myself about to 4.26/4.25 debate. I don't want to say I told you so, but I told you so.:D
Someone else with some kind of micro lags since the newest nvidia Driver Update? (4090 @ 4.15)
It feeled much smoother before the update.
SmoothVideo Project → Using SVP → New RIFE filter - 3x faster AI interpolation possible in SVP!!!
Powered by PunBB, supported by Informer Technologies, Inc.
