svr

as a quick fix you can turn on "performance boost" option in the RIFE profile

Thank you, it works!

... and maybe @Chainik can help.

Or not.

as a quick fix you can turn on "performance boost" option in the RIFE profile

svr wrote:

Hi everyone, When I try to transcode a 1080x1920 vertical video using RIFE 4.25, I get the error: “no valid optimization profile found.” How can I get SVP + RIFE to work with vertical videos?

Try to play a vertical FHD video first with RIFE activated in real time. It should build the profile before playing the video.

Real-time playback with vertical FHD and RIFE works fine without any issues. The error occurs during the transcoding process. I tried updating vsmlrt.py to the latest version, 3.22.38. That didn't help either.

08:25:48.850: ===== Starting mpv ======
08:25:48.850: Command line: C:\Program Files (x86)\SVP 4\mpv64\mpv.exe C:/Users/111/Downloads/test.mp4 --o=C:/Users/111/Downloads/test.SVP.temporary.mkv --no-audio --no-sub --no-sub-auto --input-ipc-server=mpvencodepipe --input-media-keys=no --no-msg-color --video-crop=0x0+0+0 --vf=vapoursynth:[C:\Users\111\AppData\Roaming\SVP4\scripts\ffff.py]:4:16 --of=matroska --ovc=hevc_nvenc --ovcopts=b=6201795,preset=slow,rc=vbr,maxrate=8062334,bufsize=12403590,time_base=1000/59940,threads=16
08:25:48.900: ● Video --vid=1 (h264 1080x1920 29.97 fps) [default]
08:25:48.900: ○ Audio --aid=1 --alang=eng (aac 2ch 48000 Hz 96 kbps) [default]
08:25:49.441: [vapoursynth] Script evaluation failed:
08:25:49.442: [vapoursynth] Python exception: operator (): no valid optimization profile found
08:25:49.442: [vapoursynth]
08:25:49.442: [vapoursynth] Traceback (most recent call last):
08:25:49.442: [vapoursynth] File "src\\cython\\vapoursynth.pyx", line 3387, in vapoursynth._vpy_evaluate
08:25:49.442: [vapoursynth] File "src\\cython\\vapoursynth.pyx", line 3388, in vapoursynth._vpy_evaluate
08:25:49.442: [vapoursynth] File "C:\Users\111\AppData\Roaming\SVP4\scripts\ffff.py", line 83, in <module>
08:25:49.442: [vapoursynth] smooth = interpolate(clip)
08:25:49.442: [vapoursynth] ^^^^^^^^^^^^^^^^^
08:25:49.442: [vapoursynth] File "C:\Users\111\AppData\Roaming\SVP4\scripts\ffff.py", line 61, in interpolate
08:25:49.442: [vapoursynth] smooth = RIFE_imp(input_rife,multi=rife_num,model=rife_mnum,backend=trt_backend)
08:25:49.442: [vapoursynth] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
08:25:49.442: [vapoursynth] File "C:\Program Files (x86)\SVP 4\rife\helpers.py", line 25, in RIFE_imp
08:25:49.442: [vapoursynth] return RIFE(clip,multi,1.0,None,None,None,model_num,backend,ensemble,False,implementation)
08:25:49.442: [vapoursynth] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
08:25:49.442: [vapoursynth] File "C:\Program Files (x86)\SVP 4\rife\vsmlrt.py", line 1199, in RIFE
08:25:49.442: [vapoursynth] output0 = RIFEMerge(
08:25:49.442: [vapoursynth] ^^^^^^^^^^
08:25:49.442: [vapoursynth] File "C:\Program Files (x86)\SVP 4\rife\vsmlrt.py", line 1066, in RIFEMerge
08:25:49.442: [vapoursynth] return inference_with_fallback(
08:25:49.442: [vapoursynth] ^^^^^^^^^^^^^^^^^^^^^^^^
08:25:49.442: [vapoursynth] File "C:\Program Files (x86)\SVP 4\rife\vsmlrt.py", line 2693, in inference_with_fallback
08:25:49.442: [vapoursynth] raise e
08:25:49.442: [vapoursynth] File "C:\Program Files (x86)\SVP 4\rife\vsmlrt.py", line 2670, in inference_with_fallback
08:25:49.442: [vapoursynth] ret = _inference(
08:25:49.442: [vapoursynth] ^^^^^^^^^^^
08:25:49.442: [vapoursynth] File "C:\Program Files (x86)\SVP 4\rife\vsmlrt.py", line 2570, in _inference
08:25:49.442: [vapoursynth] ret = core.trt.Model(
08:25:49.442: [vapoursynth] ^^^^^^^^^^^^^^^
08:25:49.442: [vapoursynth] File "src\\cython\\vapoursynth.pyx", line 3123, in vapoursynth.Function.__call__
08:25:49.443: [vapoursynth] vapoursynth.Error: operator (): no valid optimization profile found
08:25:49.443: [vapoursynth]
08:25:49.483: (!!!) Intermediate file may be broken: C:\Users\111\Downloads\test.SVP.temporary.mkv
08:25:49.483: ===== mpv exited with code 62097 =====

dawkinscm wrote:

Good or bad genuine feedback is useful and always welcome so thanks for the update

I have 98003d & RTX 5090, regular svp is no issue hardware wise with those overrides. Your settings seem very good for most content, might need to do a separate rife profile for harder animation. Stuck with 60hz as that is the only one my 4k dlp does well. But thanks for sharing the settings.

With that hardware you should have almost no issues with any content. I watched Super Mario up-scaled to 4K and the only difference is that I watch at 120Hz, not 60Hz. There is something I've noticed that might explain all of this and maybe @Chainik can help.

Well actually two things. First since I moved away from RIFE and started using SVP when I hover over the SVP icon in the system tray, it says "Paused" even while it's playing and clearly affecting the content. This probably ties into why the SVP Index graph shows N/A. If I reset everything the SVP Index graph starts working again for a few minutes then stops. So maybe it's the monitoring that is "paused"? The second issue might not be an issue in that SVP reports 1920x1080 for even though the file itself is actually 3840x2160. I've attached an image showing the two issues.

Good or bad genuine feedback is useful and always welcome so thanks for the update

I have 98003d & RTX 5090, regular svp is no issue hardware wise with those overrides. Your settings seem very good for most content, might need to do a separate rife profile for harder animation. Stuck with 60hz as that is the only one my 4k dlp does well. But thanks for sharing the settings.

I don't think so

just tried latest TensorRT 10.16 libs, and ... they're like 10% slower than good old 10.8 bundled with SVP, on both 4070 and 2060 cards

also tried TRT-RTX backend, which doesn't require 2+ GB of libs and with much faster engine build process
and it really works, even 10% _faster_ on 4070 (but 20% slower on 2060)
BUT it takes 15+ seconds for start-up, or after every timeline seek

Does TRT-RTX make any difference to RIFE quality and artifacts? I assume the answer is no because it is more to do with the engine and how it is trained. But it would be nice to have conformation from someone who has actually used it.

also tried TRT-RTX backend, which doesn't require 2+ GB of libs and with much faster engine build process
and it really works, even 10% _faster_ on 4070 (but 20% slower on 2060)
BUT it takes 15+ seconds for start-up, or after every timeline seek

@dawkinscm

Your settings worked very well with 60hz on film content, tried with those overrides. Unfortunately with cg animation at least they were full of artifacts. I tested with the Super Mario movie, original SVP always has issues with this one. Rife handles that one very well.

I'm glad it works well for you. Strange about Super Mario Movie as I have no issues. But that's in 120fps which as I said is paradoxically easier to smooth than 60fps. I was going to look into 60fps smooth motion but then someone essentially told me not to bother so I didn't

BTW, just a reminder that the CPU, GPU and SVP are 3 separate choke points and from the feedback I have seen so far, these settings (minus the overkills) are relatively safe and work well within those boundaries. I bring this up because I have made two small changes to my settings since I last posted. The main one is a reduction in the mv_gridsize, but I also turn Wide Search back on.  With overrides at 4k (3840x2160) resolution and 120fps, even these minor changes might stress some CPUs, kill many GPUs and are close to the SVP processing limit depending on the scene. If you want to experiment then my advice is to remove the overrides if you are using them (or at least the "overkills") ant try out different combinations of GUI settings. You might be surprised at the difference even small changes can make.

Good or bad genuine feedback is useful and always welcome so thanks for the update

Your settings worked very well with 60hz on film content, tried with those overrides. Unfortunately with cg animation at least they were full of artifacts. I tested with the Super Mario movie, original SVP always has issues with this one. Rife handles that one very well.

Hi everyone, When I try to transcode a 1080x1920 vertical video using RIFE 4.25, I get the error: “no valid optimization profile found.” How can I get SVP + RIFE to work with vertical videos?

Try to play a vertical FHD video first with RIFE activated in real time. It should build the profile before playing the video.

dawkinscm wrote:

abraxas wrote:

I'd also love to see the overrides, because the settings you just posted are already working extremely well here (120 frames per second, and the 4070 GPU is 30-40% max). The few artifacts can certainly still be "masked."

I've had a few people PM me for them and I sent them with notes.  I'm very happy that the basic settings are working well for you and it's probably all you need. With my current settings including overrides the 1% frame pacing average is around 8.4ms over wifi.  The combination 4k content on a very large screen being streamed over wifi makes even the smallest jitters more obvious which is why I monitor even the 1% averages. For most people the overrides are very overkill and no one in their right mind is monitoring frame pacing for watching movies lol. But if there is some traction for me to post them then I will.

+1 vote for you to post them, or PM them to me if you'd rather. I originally bought SVP to play with RIFE but after experimenting and reading some more I came across your basic settings and was much happier with the results. I'm using a 9700X and a 3080 Ti and have some good headroom to do 120 fps, but I've also realized the importance of keeping frametimes from getting out of hand.

Thanks. Assuming you do have the headroom then I think 120fps is actually easier to make smooth than 60fps because it is a multiple of 24fps. Below are the overrides with notes.

Remember this is for my setup. 4K streamed over WiFi at 120fps on a "very" large screen. I've labelled the ones I think might not be needed for a standard TV.

==================================
levels.pel                                    = 4;  /* Overkill */
analyse.main.search.type             = 4;
analyse.main.search.coarse.width = 3840;   /* Definitely Overkill */

/* Definitely Overkill and no longer using because it "kills" SVP when mv_gridsize is reduced */
analyse.refine[0].search.satd    = true; 

/* Overrides "Artifacts Masking" setting */
smooth.mask.cov      = 300;
smooth.mask.area     = 300;

smooth.mask.area_sharp    = 1.3; /* Personal Preference */

/* TBH I would love some help on better understanding the settings below. But from what little I do understand these work really well for me */
analyse.main.penalty.lsad        = 16000;
analyse.refine[0].penalty.lsad   = 12000;
analyse.refine[0].penalty.lambda    = 20.0;
analyse.refine[0].penalty.pnew       = 100;

abraxas wrote:

I'd also love to see the overrides, because the settings you just posted are already working extremely well here (120 frames per second, and the 4070 GPU is 30-40% max). The few artifacts can certainly still be "masked."

I've had a few people PM me for them and I sent them with notes.  I'm very happy that the basic settings are working well for you and it's probably all you need. With my current settings including overrides the 1% frame pacing average is around 8.4ms over wifi.  The combination 4k content on a very large screen being streamed over wifi makes even the smallest jitters more obvious which is why I monitor even the 1% averages. For most people the overrides are very overkill and no one in their right mind is monitoring frame pacing for watching movies lol. But if there is some traction for me to post them then I will.

+1 vote for you to post them, or PM them to me if you'd rather. I originally bought SVP to play with RIFE but after experimenting and reading some more I came across your basic settings and was much happier with the results. I'm using a 9700X and a 3080 Ti and have some good headroom to do 120 fps, but I've also realized the importance of keeping frametimes from getting out of hand.