New RIFE filter - 3x faster AI interpolation possible in SVP!!! (Page 45 of 84)

Hi, I was trying to use Rife in the SVP. I've been only watching anime using SVP with non-RIFE settings since I purchased it which is 3 years ago.
I am totally clueless about these RIFE feature and tried to do my own research and test but was stuck at one thing, which shown as below:

The error "Your ONNX Model has been generated with INT64 weights, while TensorRT does not natively support INT64. Attempting to cast down to INT32."

After this message is shown, my playback will be stuck and only audio plays, sometimes even audio didn't play.

This is my setting for the RIFE

(that custom SC Threshold is because i tried following a youtube tutorial, but it still produced the same thing. Maybe I am doing it wrong)

I know there's probably an answer for this question already somewhere, I tried searching for it but there's too many too unrelated (probably) topic around it. a
I hope someone can guide me in getting Soap Opera effect using RIFE in SVP.
Sorry for the troubles and thanks for any consideration given.

-----------------------------------------------------------
Note that my pc setup is as such:

- Monitor 24" 144Hz
- AMD Ryzen 5 3600
- RTX 3080Ti 12GB
- 32GB DDR4 3200GHz ram
- around 500GB+ free space in the 1TB Adata S70 Blade M.2 NVme SSD

---------------------------------------------------------------
EDIT :

Nevermind, I am just too stupid. All i need to do is wait a bit longer and it'll works (using the settings in the picture)
But its weird, i swear i waited close to 30 mins before, hence I was thinking it's actually stuck/crashed. Now I just need to wait for 5 min at most.

Using MPC-HC since it has built-in fps counter. Would love to use MPV but I can't find the way to enable fps counter.

flowreen91 wrote:

hvs wrote:

Would love to use MPV but I can't find the way to enable fps counter.

Try Shift + I or hold I to enable fps counter.
MPV seems faster than MPC-HC. It gives more stable fps.

hvs wrote:

I hope someone can guide me in getting Soap Opera effect using RIFE in SVP.

Try this and say if you get stuck or instructions are unclear :
https://www.svp-team.com/forum/viewtopi … 352#p83352

Thanks you so much for this. Yes it's true that MPV is much more stable.

aloola wrote:

hvs wrote:

I hope someone can guide me in getting Soap Opera effect using RIFE in SVP.

currently, the best quality that almost gets rid of the soap effect is models 4.12 and 4.13.

4.0->4.6: about the same quality and same performance
4.7->4.9: about the same quality and same performance, the performance cost +20% than 4.0->4.6
4.10->4.13: about the same quality and same performance, the performance cost +20% than 4.7->4.9
4.12lite, 4.13lite: performance between 4.7->4.9 and 4.10->4.13

my recommended models are: 4.6, 4.9 and 4.13

Thanks for the recommendation. When searching the solution for my problem, I read in this post regarding how v2 has much more advantage than v1..
So I tried 4.9 v2... but it seems like there's even Ensemble version, which I didn't find anywhere about it said here, or probably just missed the conversation regarding it.

4.9 V2 seems very stable for me, and the wait isnt that long either, i just have to wait for no more than 1 minute or two.
Tried to encode Kagejitsu S2 Ep 12 using it before i went to sleep too, but iirc it finished around 3 hours (because my pc auto shut down around that time).
The result is so great. the artifacts are almost none, and even if its there, its very subtle, which makes it worth the wait than watching the anime with live interpolation. This is a good option for me if I had no need to watch the anime ASAP.

All those talk aside, what I actually want to achieve is that soap effect that everyone hates. I am weird in that I love that soap effect XD.
I'll try 4.13 later I guess.

RickyAstle98 wrote:

Idk why you wait 5 minutes, maybe engine build is CPU limited? I have 10700K and RTX 4070 and NVMe too, but slower cache build time was 2m24sec only!

eh that was just my estimate for max time i waited (not timed).. its actually less than that ig. mostly around 30-1 mins. Ig it depends on the scene to interpolate

---------------
Also talking about CPU..
Can't we use GPU as a main to transcode? or is it only just mainly CPU with GPU as backup?

hvs wrote:

flowreen91 wrote:

Try Shift + I or hold I to enable fps counter.
MPV seems faster than MPC-HC. It gives more stable fps.

Try this and say if you get stuck or instructions are unclear :
https://www.svp-team.com/forum/viewtopi … 352#p83352

Thanks you so much for this. Yes it's true that MPV is much more stable.

aloola wrote:

currently, the best quality that almost gets rid of the soap effect is models 4.12 and 4.13.

4.0->4.6: about the same quality and same performance
4.7->4.9: about the same quality and same performance, the performance cost +20% than 4.0->4.6
4.10->4.13: about the same quality and same performance, the performance cost +20% than 4.7->4.9
4.12lite, 4.13lite: performance between 4.7->4.9 and 4.10->4.13

my recommended models are: 4.6, 4.9 and 4.13

Thanks for the recommendation. When searching the solution for my problem, I read in this post regarding how v2 has much more advantage than v1..
So I tried 4.9 v2... but it seems like there's even Ensemble version, which I didn't find anywhere about it said here, or probably just missed the conversation regarding it.

4.9 V2 seems very stable for me, and the wait isnt that long either, i just have to wait for no more than 1 minute or two.
Tried to encode Kagejitsu S2 Ep 12 using it before i went to sleep too, but iirc it finished around 3 hours (because my pc auto shut down around that time).
The result is so great. the artifacts are almost none, and even if its there, its very subtle, which makes it worth the wait than watching the anime with live interpolation. This is a good option for me if I had no need to watch the anime ASAP.

All those talk aside, what I actually want to achieve is that soap effect that everyone hates. I am weird in that I love that soap effect XD.
I'll try 4.13 later I guess.

RickyAstle98 wrote:

Idk why you wait 5 minutes, maybe engine build is CPU limited? I have 10700K and RTX 4070 and NVMe too, but slower cache build time was 2m24sec only!

eh that was just my estimate for max time i waited (not timed).. its actually less than that ig. mostly around 30-1 mins. Ig it depends on the scene to interpolate

---------------
Also talking about CPU..
Can't we use GPU as a main to transcode? or is it only just mainly CPU with GPU as backup?

As I can say, I test 720p30>720p90 transcode, SVP window show 25% all the time, but my CPU maxed at 7% only during transcoding, NVENC encoder!
So its GPU mostly transcode...

Decided to test 4.13 and 4.13 v2.

4.13 works well, Win resource monitor shows ~50% GPU utilization regardless of the video resolution. I tested with 720p and 1080p videos at 60 Hz. Shouldn't there be some sort of difference between 720p and 1080p? 720p even reaches 60% more often on my RTX 4080. Also, the 4.9 model included with SPV by default now has almost the same performance, mostly sitting at ~50% regardless of resolution.

4.13 v2 doesn't seem to be working, however. Every time I try to play a video, it compiles the cache file and after around 50 seconds it stops and the video plays without RIFE. Can't really see if it gives an error cuz the window closes fast.

RickyAstle98 wrote:

Xenocyde wrote:

Decided to test 4.13 and 4.13 v2.

4.13 works well, Win resource monitor shows ~50% GPU utilization regardless of the video resolution. I tested with 720p and 1080p videos at 60 Hz. Shouldn't there be some sort of difference between 720p and 1080p? 720p even reaches 60% more often on my RTX 4080. Also, the 4.9 model included with SPV by default now has almost the same performance, mostly sitting at ~50% regardless of resolution.

4.13 v2 doesn't seem to be working, however. Every time I try to play a video, it compiles the cache file and after around 50 seconds it stops and the video plays without RIFE. Can't really see if it gives an error cuz the window closes fast.

Something is wrong bruh >
I have 10700K and RTX 4070 and here is my performance!
4.4v2 720p24>144 (<45% GPU usage) at 165Hz
4.12v2 720p24>144 (<65% GPU usage) at 165Hz
4.13v2 720p24>144 (<70% GPU usage) at 165Hz

I only test with 60 Hz cuz my screen is old af Have you tested with 60 Hz? I feel like the utilization might not change too much even if refresh is X4+ for me.

dawkinscm wrote:

Make sure you have a 4.13 onnx file in the "v2" folder. If you do then delete it and any files created by it, download the v2 onnx file again and put the new onnx file into the "v2" folder.

Deleted, redownlaoded, still doesn't work. Do I need some new version of Python or TRT maybe?

Wait me here, I will edit my message in next 5 minutes! Check plz!!!
So you need  to put 4.13.onnx into v2 folder where SVP located
https://ibb.co.com/RYCzGRh
If your 1st timing cache build failed, you need enable TRT boost option, 4.7-4.9v2 cant stable buld timing cache for general optimization... Maybe 4.13v2 too? Me too cant build timing cache without TRT enabled!
Yes I tested 60Hz and only CPU usage change a little, anyway I have modern 165Hz GSync monitor, any FPS factor, smooth always!

Hehe))

V2 models for some reason have jiggling, the more to the right and down, the worse the effect. Almost unnoticeable for 1080p, but very annoying for 1440p and 4K (both native and downscaled to 1440p, here's the example - https://cdn.discordapp.com/attachments/ … WGll8w.mp4). For this one, I'm running Game of Thrones S1 E1 at 4K on 3440x1440 (so SVP downscales to 2560x1440) at x2 framerate on 2080 Ti with 4.4v2, 4.6v2, 4.9v2 and 4.13_litev2. There's no such effect on V1 models (tried all same versions)