1,901 (edited by Blackfyre 04-09-2024 13:49:29)

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

MAG79 wrote:

Blackfyre
I believe you have compared 48 Hz custom mode and 50 Hz speedup mode. What do you think?

Oh I haven't tried speed up since I switched to RIFE, I did before RIFE and it was fine. I might give it a try, but I am very happy with 48FPS @ 48Hz

Chainik wrote:

trying to figure the best default model now
and it looks like 4.6 is still the best default choice

100% keep 4.6 as the default for everyone. Because then people will complain all of a sudden their RIFE is lagging if it changes the default with an update haha!

Most 'advanced' users already know how to change them.

Maybe you can add the 'extra options'. Such as Rife 4.16 Lite v2 and 4.18 in the drop down menu with 4.9 as well, but keep 4.6 as default.

Chainik wrote:

ok, 4.18 is also two times slower, and it's also recommended by the model creator
is it two times better?

Is it two times better? It depends, I have always been sensitive to artifacts, since old SVP days I used to always try to minimise artifacts haha

So to some of us, it is a lot more than 2x better.

In terms of 'smoothness', not much difference, very similar between all RIFE versions. In terms of artifact masking, much much better.

1,902

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Hey. When can we get RIFE working on Archlinux or Linux in general?

1,903

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

SVP packages updated: TensortRT 10.4.0, vsmlrt 15.4

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Chainik wrote:

SVP packages updated: TensortRT 10.4.0, vsmlrt 15.4

Thank you. I probably wasn't going to update to 10.4,. But I'm glad that Cuda 12.6 the  officially part of SVP. Thanks again smile

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Wasn't this a great time to fully switch to the v2 models?
Or are there certain scenarios where they actually perform slower than v1 models?

1,906 (edited by oriento 11-09-2024 08:43:23)

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

thanks for the update. I've tested and I don't see any performance gain or lose with rife 4.22 lite v2. Around 98fps for converting a specific video 1080

1,907

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

4.18 (v2) still the best atm?

1,908 (edited by flowreen91 12-09-2024 07:31:03)

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

John86 wrote:

4.18 (v2) still the best atm?

On RIFE page: https://github.com/hzwer/Practical-RIFE
RIFE devs say "For anime scenes, it is recommended to choose 4.22.lite, and for real scenes, it is recommended to choose 4.18."
So i would go with 4.22_lite v2 cause it's fastest recent lower computational cost model.

1,909 (edited by Blackfyre 13-09-2024 13:51:45)

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Chainik wrote:

SVP packages updated: TensortRT 10.4.0, vsmlrt 15.4

Edit: Making corrections as my previous statement was false in regards to performance.

There is still a performance difference between v1 and v2, and v2 still performs better, but the gap is smaller than what it was before. Good update overall, but it reintroduced an old v2 problem I had previously solved.

For those who are using v2. If you notice screen shaking in some scenes, or on certain elements on screen only. The fix is simple and I will post it below.

If you do not notice this screen shaking with certain elements or background items in films or shows, then do not worry about this, and this ONLY applies to v2 and advanced users who mess around with RIFE.

NOT for default users (you will not experience these issues).

Now for the fix.

Go to the link below, scroll down, download scripts.v15.4.7z (current latest), extract and copy vsmlrt.py and place it in the C:\Program Files (x86)\SVP 4\rife folder and overwrite the one that comes with SVP.

The updated script should fix the screen shake problem:

https://github.com/AmusementClub/vs-mlrt/releases

Thanks to flowreen91 for identifying the problem 20 or 30 pages back and posting the solution.

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Updated comment above multiple times for accuracy. Also thanks to flowreen91 for posting the original fix for screen shaking in some scenarios while using v2's

1,911

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

I believe that all "fixes" the latest vsmlrt.py contains are included in SVP's one

1,912 (edited by flowreen91 13-09-2024 17:12:08)

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Blackfyre wrote:

Also thanks to flowreen91 for posting the original fix for screen shaking in some scenarios while using v2's

The old tree wiggle issue was present only on TensorRT v14.test2 and older, and was fixed just by upgrading to TensorRT 14.test3 or newer.
https://www.svp-team.com/forum/viewtopi … 617#p84617
It was unrelated to what's inside the vsmlrt.py file.

But if u indeed manage to reproduce a new screen shaking issue and fix it on your 4K screen while using latest TensorRT please provide more data/video/recording example so Chainik can look into it. ^_^

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Chainik wrote:

I believe that all "fixes" the latest vsmlrt.py contains are included in SVP's one

Perhaps someone else can test and see to confirm v2 screen shake issue. Solved by updating to the latest vsmlrt.py

Testing was done with 4.16 Lite (v2).

First episode of Ahsoka below is very easy to notice, around minute 2 and 17 seconds, the highlighted area in the image below (vertical and horizontal neon lights) shake. The ones on the left side of the image do not shake.

Ahsoka.S01E01.HDR.2160p.WEB.H265

https://i.imgur.com/WTiBkqw.jpeg

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

flowreen91 wrote:
Blackfyre wrote:

Also thanks to flowreen91 for posting the original fix for screen shaking in some scenarios while using v2's

The old tree wiggle issue was present only on TensorRT v14.test2 and older, and was fixed just by upgrading to TensorRT 14.test3 or newer.
https://www.svp-team.com/forum/viewtopi … 617#p84617
It was unrelated to what's inside the vsmlrt.py file.

But if u indeed manage to reproduce a new screen shaking issue and fix it on your 4K screen while using latest TensorRT please provide more data/video/recording example so Chainik can look into it. ^_^

Maybe because I had updated stuff manually before, SVP failed to update some sections (such as the latest vsmlrt.py) when it done the other updates.

I will be formatting my computer soon, so I can test on a full fresh install as well.

I guess for now, discard everything I said above until I test on a new setup. Or someone else can confirm the issue with the example I listed above.

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Blackfyre wrote:

I guess for now, discard everything I said above until I test on a new setup.

Formatted, clean Windows install. Clean SVP install.

Can confirm 100% the shake issue is there for v2 until I update vsmlrt.py which solves the problem as outlined above.

1,916 (edited by flowreen91 15-09-2024 08:01:35)

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Chainik wrote:

I believe that all "fixes" the latest vsmlrt.py contains are included in SVP's one

The v2 shake issue seems to get fixed if u add this on line 1027:
    video_player: bool = False,
https://gyazo.com/d7798e1ee97ecfa658d8587466561630
(nevermind it seems it just makes it load the v1 model instead)

1,917 (edited by dawkinscm 15-09-2024 06:57:50)

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

flowreen91 wrote:
Chainik wrote:

I believe that all "fixes" the latest vsmlrt.py contains are included in SVP's one

The v2 shake issue seems to get fixed if u add this on line 1027:
    video_player: bool = False,
https://gyazo.com/d7798e1ee97ecfa658d8587466561630
Please take a look.

I think that should be on by default because from what I have read, it might be duplicating some SVP functionality which may explain why the shaking bug occurs.

1,918 (edited by TechnoStone 15-09-2024 02:08:02)

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Just installed latest update of SVP, and it no longer sees the v2 model:

01:07:47.382 [i]: Playback: starting up...
01:07:47.386 [i]: Playback [530c1dc6]: Frame server (64-bit) C:\Program Files (x86)\SVP 4\mpv64\vapoursynth.dll
01:07:47.388 [i]: Playback [530c1dc6]: resulting video frame 1920x1080
01:07:47.390 [i]: Playback [530c1dc6]: 3 acceptible profiles, best is 'RIFE AI engine' [100]
01:07:47.399 [i]: Playback [530c1dc6]: enabled while video is playing
01:07:47.403 [i]: Playback [530c1dc6]: playing at 71.928 [23.976 *3/1] 
01:07:47.761 [E]: Playback [530c1dc6]: VS - could not init VS
01:07:47.765 [E]: Playback [530c1dc6]: VS - Script evaluation failed:
01:07:47.765 [E]: Playback [530c1dc6]: VS - Python exception: 'C:/Program Files (x86)/SVP 4/rife\models\rife\rife_v4.22_lite.onnx' not found, built-in models can be found at https://github.com/AmusementClub/vs-mlrt/releases/tag/model-20211209, https://github.com/AmusementClub/vs-mlrt/releases/tag/model-20220923 and https://github.com/AmusementClub/vs-mlrt/releases/tag/external-models
01:07:47.765 [E]: Playback [530c1dc6]: VS - Traceback (most recent call last):
01:07:47.765 [E]: Playback [530c1dc6]: VS - File 'src\\cython\\vapoursynth.pyx', line 3121, in vapoursynth._vpy_evaluate
01:07:47.765 [E]: Playback [530c1dc6]: VS - File 'src\\cython\\vapoursynth.pyx', line 3122, in vapoursynth._vpy_evaluate
01:07:47.765 [E]: Playback [530c1dc6]: VS - File 'C:\Users\TechnoStone\AppData\Roaming\SVP4\scripts\530c1dc6.py', line 84, in <module>
01:07:47.765 [E]: Playback [530c1dc6]: VS - smooth = interpolate(clip)
01:07:47.765 [E]: Playback [530c1dc6]: VS - ^^^^^^^^^^^^^^^^^
01:07:47.765 [E]: Playback [530c1dc6]: VS - File 'C:\Users\TechnoStone\AppData\Roaming\SVP4\scripts\530c1dc6.py', line 62, in interpolate
01:07:47.765 [E]: Playback [530c1dc6]: VS - smooth = RIFE_imp(input_rife,multi=rife_num,model=rife_mnum,backend=trt_backend)
01:07:47.765 [E]: Playback [530c1dc6]: VS - ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
01:07:47.766 [E]: Playback [530c1dc6]: VS - File 'C:\Program Files (x86)\SVP 4\rife\helpers.py', line 23, in RIFE_imp
01:07:47.766 [E]: Playback [530c1dc6]: VS - return RIFE(clip,multi,1.0,None,None,None,model_num,backend,ensemble,implementation)
01:07:47.766 [E]: Playback [530c1dc6]: VS - ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
01:07:47.766 [E]: Playback [530c1dc6]: VS - File 'C:\Program Files (x86)\SVP 4\rife\vsmlrt.py', line 1197, in RIFE
01:07:47.766 [E]: Playback [530c1dc6]: VS - if isinstance(max_shapes, int):
01:07:47.766 [E]: Playback [530c1dc6]: VS - ^^^^^^^^^^^^^^^^^
01:07:47.766 [E]: Playback [530c1dc6]: VS - File 'C:\Program Files (x86)\SVP 4\rife\vsmlrt.py', line 1064, in RIFEMerge
01:07:47.766 [E]: Playback [530c1dc6]: VS - raise ValueError(f'{func_name}: RIFE: multi must be at least 2')
01:07:47.766 [E]: Playback [530c1dc6]: VS - ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
01:07:47.766 [E]: Playback [530c1dc6]: VS - File 'C:\Program Files (x86)\SVP 4\rife\vsmlrt.py', line 2582, in inference_with_fallback
01:07:47.766 [E]: Playback [530c1dc6]: VS - File 'C:\Program Files (x86)\SVP 4\rife\vsmlrt.py', line 2561, in inference_with_fallback
01:07:47.766 [E]: Playback [530c1dc6]: VS - File 'C:\Program Files (x86)\SVP 4\rife\vsmlrt.py', line 2325, in _inference
01:07:47.766 [E]: Playback [530c1dc6]: VS - RuntimeError: 'C:/Program Files (x86)/SVP 4/rife\models\rife\rife_v4.22_lite.onnx' not found, built-in models can be found at https://github.com/AmusementClub/vs-mlrt/releases/tag/model-20211209, https://github.com/AmusementClub/vs-mlrt/releases/tag/model-20220923 and https://github.com/AmusementClub/vs-mlrt/releases/tag/external-models

https://files.catbox.moe/ovjg6c.pnghttps://files.catbox.moe/ocpo95.png
Any ways to fix? UPD. see next post

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Blackfyre wrote:
Blackfyre wrote:

I guess for now, discard everything I said above until I test on a new setup.

Formatted, clean Windows install. Clean SVP install.
Can confirm 100% the shake issue is there for v2 until I update vsmlrt.py which solves the problem as outlined above.

flowreen91 wrote:
Chainik wrote:

I believe that all "fixes" the latest vsmlrt.py contains are included in SVP's one

The v2 shake issue seems to get fixed if u add this on line 1027:
    video_player: bool = False,
https://gyazo.com/d7798e1ee97ecfa658d8587466561630
Please take a look.

Messed with reinstallations, BOTH of these methods just make svp use v1 models, which doesn't have shaking issue and this caused the issue I've described above (it just can't load missing model for sure, I just never use v1 models so I don't even unpack them). For some reason, my script file wasn't correctly updated and partial reinstall (via components) couldn't bring me a working variant

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

dawkinscm wrote:
flowreen91 wrote:
Chainik wrote:

I believe that all "fixes" the latest vsmlrt.py contains are included in SVP's one

The v2 shake issue seems to get fixed if u add this on line 1027:
    video_player: bool = False,
https://gyazo.com/d7798e1ee97ecfa658d8587466561630
Please take a look.

I think that should be on by default because from what I have read, it might be duplicating some SVP functionality which may explain why the shaking bug occurs.

To clarify, Setting to "True" might duplicate some SVP functionality. I meant' that video_player: bool = False should be the default setting.

1,921 (edited by flowreen91 15-09-2024 07:34:38)

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

TechnoStone wrote:

Messed with reinstallations, BOTH of these methods just make svp use v1 models

Hahaha you are right! I never noticed that.
It means the fix is actually longer than one line, hidden somewhere in the vsmlrt.py file from scripts.v15.4.7z
Will wait for SVP developers for actual fix.
Thanks TechnoStone!

Temporarily we can use these that don't have the v2 shake and load v2 models correctly:
Attaching latest version of vsmlrt.py from here https://github.com/AmusementClub/vs-mlr … /vsmlrt.py

dawkinscm wrote:

video_player: bool = False should be the default setting.

Attaching sending video_player False to RIFE as dawkinscm suggests in helpers.py

Post's attachments

helpers.py 717 b, 49 downloads since 2024-09-15 

vsmlrt.py 102.26 kb, 32 downloads since 2024-09-15 

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

This explains why I was seeing equal performance before.

To confirm the discussion above, I went to the rife folder and deleted all the models inside rife, and only kept the ones inside rife_v2

RIFE is not working without the v1 models, so basically it means it was just using v1 models even when v2 was selected?

flowreen91 wrote:

Attaching sending video_player False to RIFE as dawkinscm suggests in helpers.py

Tested with these above and also rife v2 is not working.

To test properly, remove all the models inside v1, so SVP doesn't fall back to them. That's what I am doing.

Now I need to find a way to make v2 work.

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Blackfyre wrote:

This explains why I was seeing equal performance before.

To confirm the discussion above, I went to the rife folder and deleted all the models inside rife, and only kept the ones inside rife_v2

RIFE is not working without the v1 models, so basically it means it was just using v1 models even when v2 was selected?

flowreen91 wrote:

Attaching sending video_player False to RIFE as dawkinscm suggests in helpers.py

Tested with these above and also rife v2 is not working.

To test properly, remove all the models inside v1, so SVP doesn't fall back to them. That's what I am doing.

Now I need to find a way to make v2 work.

I think I remember Chainik saying that v2 models have become pointless. Especially after the change (can't remember the name, it's a Windows dll) that made it work better. But if you wish to process v2 files then maybe you should compare the helper files from before and after the latest update. From memory, the v2 processing is technically in the vslmt script, but I think the SVP devs essentially moved that functionality to the helper script. But as I said this is from memory.

1,924 (edited by flowreen91 15-09-2024 10:44:52)

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

Blackfyre wrote:

To test properly, remove all the models inside v1, so SVP doesn't fall back to them. That's what I am doing.

Yup doing that and works for me with files attached above:
https://gyazo.com/a8b01ddcff578c3033de7b50522481a1
Try to delete the old cache folder from here "C:\Users\Blackfyre\AppData\Roaming\SVP4\cache" and restart SVP maybe?

Re: New RIFE filter - 3x faster AI interpolation possible in SVP!!!

flowreen91 wrote:
Blackfyre wrote:

To test properly, remove all the models inside v1, so SVP doesn't fall back to them. That's what I am doing.

Yup doing that and works for me with files attached above:
https://gyazo.com/a8b01ddcff578c3033de7b50522481a1
Try to delete the old cache folder from here "C:\Users\Blackfyre\AppData\Roaming\SVP4\cache" and restart SVP maybe?

Files attached above are working correctly now after restarting the PC.

Closing SVP and opening it before was not working.

I'll delete all the cache files too

I will put back v1 models and test performance difference now between v1 and v2 properly.