1,826 Reply by Xenocyde

  • Thanks: 43
  • Thanks for the post: 1

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

dawkinscm wrote:
Xenocyde wrote:
flowreen91 wrote:

The devs fixed the smoothness issue:
https://github.com/hzwer/Practical-RIFE … 2275333756

I haven't noticed any smoothness problems since I don't watch anime. I only tested with my usual football scenes
and 4.21 still has the same edge artifacts as 4.20 in those scenes.

BTW, 4.22 is out. Maybe they fixed it with this one. Testing now.

4.22 is the fix he's talking about. It's a little better than 4.21 but overall it has the same artefacts.

Ah thought it was a 4.21 update, they don't actually mention 4.22 in the comments.

  • Thanks: 176
  • Thanks for the post: 1

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

Xenocyde wrote:
dawkinscm wrote:
Xenocyde wrote:

I haven't noticed any smoothness problems since I don't watch anime. I only tested with my usual football scenes
and 4.21 still has the same edge artifacts as 4.20 in those scenes.

BTW, 4.22 is out. Maybe they fixed it with this one. Testing now.

4.22 is the fix he's talking about. It's a little better than 4.21 but overall it has the same artefacts.

Ah thought it was a 4.21 update, they don't actually mention 4.22 in the comments.

The dev mentioned it.

1,828 Reply by Xenocyde

  • Thanks: 43
  • Thanks for the post: 1

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

From my quick test I can see that 4.22 fixed the edge artifacts and scene transitions seem better compared to 4.18, but interference patterns are a bit more obvious. I'll watch an entire episode of Dr. Who later to see if there are other problems sticking out.

1,829 Reply by dawkinscm (edited by dawkinscm 08-08-2024 13:03:22)

  • Thanks: 176
  • Thanks for the post: 1

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

I assume that SVP motion detection is the default now when using SVP and I would like to use it if it is useful. But is it possible to fix the garbage frame issue where it and NVOF behave identical to "disable" motion detection? As shown before, IC is still the only motion detect that behaves properly.

1,830 Reply by erriep2703 (edited by erriep2703 08-08-2024 19:27:08)

  • Thanks: 1
  • Thanks for the post: 1

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

What is the best Anime Setup?
Which Rife Version, mpv config, svp config, etc.? 

At the moment i use:
https://www.svp-team.com/forum/viewtopi … 352#p83352
with Rife 4.22(v2) and newest TensorTR library -> v15.2.

My MPV Conf:

#
# THIS FILE WILL BE OVERWRITTEN WITH THE PACKAGE UPDATE
# Use %APPDATA%/mpv/mpv.conf
# OR SVP's menu -> Information -> Add. information -> mpv config. file
#     to define your own customized configuration
#
# Sample mpv configuration
#

# this is required for SVP to "catch" the mpv
input-ipc-server=mpvpipe

# hardware video decoder
#hwdec=auto-copy
hwdec-codecs=all
hr-seek-framedrop=no
no-resume-playback

vo=gpu-next

# high quality video output, require rather fast video card
profile=gpu-hq

# D3D11 renderer (default) is required for the HDR playback
gpu-api=d3d11

# use VULKAN API if you have problems with the default d3d11 renderer
#gpu-api=vulkan

# miscellaneous options

# comment this one if you want mpv to remember the playback position
save-position-on-quit=yes

# fixes audio desync
hr-seek-framedrop=no

# "ReClock" replacement
#video-sync=display-resample
#video-sync-max-video-change=5

#log-file=mpv.log

# Deband filter. Always turn on for anime.
deband=yes # Default values are 1:64:16:48

# Deband parameters configuration.
deband-iterations=2 # Range 1-16.
deband-threshold=35 # Range 0-4096.
deband-range=20 # Range 1-64.
deband-grain=5 # Range 0-4096.

dither-depth=auto

# Optimized shaders for higher-end GPU: Mode A (HQ)
glsl-shaders="~~/shaders/Anime4K_Clamp_Highlights.glsl;~~/shaders/Anime4K_Restore_CNN_VL.glsl;~~/shaders/Anime4K_Upscale_CNN_x2_VL.glsl;~~/shaders/Anime4K_AutoDownscalePre_x2.glsl;~~/shaders/Anime4K_AutoDownscalePre_x4.glsl;~~/shaders/Anime4K_Upscale_CNN_x2_M.glsl"

What are your favorite MPV scripts?


My SVP settings:

Post's attachments

Screenshot 2024-08-08 212243.png, 53.18 kb, 601 x 732
Screenshot 2024-08-08 212243.png 53.18 kb, 42 downloads since 2024-08-08 

  • Thanks: 74
  • Thanks for the post: 2

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

erriep2703 wrote:

# Deband filter. Always turn on for anime.
deband=yes # Default values are 1:64:16:48

I see that these settings are present in an outdated guide:
https://kokomins.wordpress.com/2019/10/ … fig-guide/
<meta property="article:modified_time" content="2024-02-08T06:53:17+00:00"/>

Probably we should use more recent configs from 2024:
https://kohana.fi/article/mpv-for-anime
https://artoriuz.github.io/blog/mpv_upscaling.html
https://iamscum.wordpress.com/guides/vi … /mpv-conf/
https://thewiki.moe/tutorials/mpv/

1,832 Reply by dawkinscm (edited by dawkinscm 08-08-2024 23:24:21)

  • Thanks: 176
  • Thanks for the post: 3

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

flowreen91 wrote:
erriep2703 wrote:

# Deband filter. Always turn on for anime.
deband=yes # Default values are 1:64:16:48

I see that these settings are present in an outdated guide:
https://kokomins.wordpress.com/2019/10/ … fig-guide/
<meta property="article:modified_time" content="2024-02-08T06:53:17+00:00"/>

Probably we should use more recent configs from 2024:
https://kohana.fi/article/mpv-for-anime
https://artoriuz.github.io/blog/mpv_upscaling.html
https://iamscum.wordpress.com/guides/vi … /mpv-conf/
https://thewiki.moe/tutorials/mpv/

I know nothing about Anime shaders so I will leave that to the experts smile  But there's nothing wrong with his config as such, he's just using the default SVP mpv.conf. Yes the config is a little outdated but there's nothing there that will make his output look worse. Then he adds some custom deband settings which is personal preference.

1,833 Reply by Insindro (edited by Insindro 17-08-2024 16:53:06)

  • Thanks: 3
  • Thanks for the post: 1

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

Can anyone tell me if there’s a way to add a delay to svp’s injection time?

1,834 Reply by RickyAstle98 (edited by RickyAstle98 09-08-2024 10:40:11)

  • Thanks: 48

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

New TensorRT library!
https://github.com/AmusementClub/vs-mlr … a.v15.2.7z

1,835 Reply by Xenocyde

  • Thanks: 43

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

RickyAstle98 wrote:

New TensorRT library!
https://github.com/AmusementClub/vs-mlr … a.v15.2.7z

Worth installing over 9.2?

1,836 Reply by RickyAstle98 (edited by RickyAstle98 09-08-2024 11:30:39)

  • Thanks: 48

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

Xenocyde wrote:
RickyAstle98 wrote:

New TensorRT library!
https://github.com/AmusementClub/vs-mlr … a.v15.2.7z

Worth installing over 9.2?

According to my realtime tests and encoding tests, the difference is starting to increase from 4.16 models!
The old models like 4.4 (v1/v2) encoding speed with 2% performance difference and maximum to 12% for newest models!
Fixed my issues? Nope, but now models look smoother a little bit! Seek performance? Variable!
Worth? If you want, you can test, for me new models looks smoother, what happens to you - who knows?!

1,837 Reply by dawkinscm (edited by dawkinscm 09-08-2024 11:57:19)

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

RickyAstle98 wrote:
Xenocyde wrote:
RickyAstle98 wrote:

New TensorRT library!
https://github.com/AmusementClub/vs-mlr … a.v15.2.7z

Worth installing over 9.2?

According to my realtime tests and encoding tests, the difference is starting to increase from 4.16 models!
The old models like 4.4 (v1/v2) encoding speed with 2% performance difference and maximum to 12% for newest models!
Fixed my issues? Nope, but now models look smoother a little bit! Seek performance? Variable!
Worth? If you want, you can test, for me new models looks smoother, what happens to you - who knows?!

@aloola already mentioned TRT 15.2 a couple of days ago and as I said at the time, on initial testing, I didn't see anything obviously better when 9.2. But I like finally having libraries that are closer to those used in the latest Nvidia drivers.

1,838 Reply by flowreen91 (edited by flowreen91 09-08-2024 12:19:13)

  • Thanks: 74

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

aloola wrote:

v15.2 gives me a tiny FPS boost
v14test3 ~150fps vs ~ v15.2 ~158fps, also fixed dynamic shape (Performance boost: off) bug

15.2 changelog:

    Upgraded to TensorRT 10.3.0.

    Fixed performance regression of RIFE and SAFA models starting with vs-mlrt v14.test4. This version may still be slightly slower than vs-mlrt v14.test3 under some conditions, however.

Looking forward to @aloola benchmarks for the new TensorRT like he did before:
https://www.svp-team.com/forum/viewtopi … 324#p84324

1,839 Reply by dawkinscm (edited by dawkinscm 09-08-2024 13:44:24)

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

flowreen91 wrote:
aloola wrote:

v15.2 gives me a tiny FPS boost
v14test3 ~150fps vs ~ v15.2 ~158fps, also fixed dynamic shape (Performance boost: off) bug

15.2 changelog:

    Upgraded to TensorRT 10.3.0.

    Fixed performance regression of RIFE and SAFA models starting with vs-mlrt v14.test4. This version may still be slightly slower than vs-mlrt v14.test3 under some conditions, however.

Looking forward to @aloola benchmarks for the new TensorRT like he did before:
https://www.svp-team.com/forum/viewtopi … 324#p84324

As he said above, on average he got around an 8 fps improvement. But this stat should be taken in the context of dev notes that suggest there might be be specific situations where the performance might be slightly worse.

1,840 Reply by RickyAstle98 (edited by RickyAstle98 09-08-2024 15:06:28)

  • Thanks: 48
  • Thanks for the post: 1

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

dawkinscm wrote:
flowreen91 wrote:
aloola wrote:

v15.2 gives me a tiny FPS boost
v14test3 ~150fps vs ~ v15.2 ~158fps, also fixed dynamic shape (Performance boost: off) bug

15.2 changelog:

    Upgraded to TensorRT 10.3.0.

    Fixed performance regression of RIFE and SAFA models starting with vs-mlrt v14.test4. This version may still be slightly slower than vs-mlrt v14.test3 under some conditions, however.

Looking forward to @aloola benchmarks for the new TensorRT like he did before:
https://www.svp-team.com/forum/viewtopi … 324#p84324

As he said above, on average he got around an 8 fps improvement. But this stat should be taken in the context of dev notes that suggest there might be be specific situations where the performance might be slightly worse.

My actual scenarios
v14test3 320fps / v15.2 310fps (4.4v2 with 1.78 dynamic shape and 320 fps with 2.35 dynamic shape)
Starting models 4.16 everything gets worse, just a performance drops variables from 6 to 12% respectively, the movie dynamic shapes always faster (even when the targeted resolution is slightly higher)

1,841 Reply by dawkinscm (edited by dawkinscm 09-08-2024 16:24:35)

  • Thanks: 176
  • Thanks for the post: 1

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

RickyAstle98 wrote:
dawkinscm wrote:
flowreen91 wrote:

Looking forward to @aloola benchmarks for the new TensorRT like he did before:
https://www.svp-team.com/forum/viewtopi … 324#p84324

As he said above, on average he got around an 8 fps improvement. But this stat should be taken in the context of dev notes that suggest there might be be specific situations where the performance might be slightly worse.

My actual scenarios
v14test3 320fps / v15.2 310fps (4.4v2 with 1.78 dynamic shape and 320 fps with 2.35 dynamic shape)
Starting models 4.16 everything gets worse, just a performance drops variables from 6 to 12% respectively, the movie dynamic shapes always faster (even when the targeted resolution is slightly higher)

So it works better with Performance Boost disabled? Am I understanding this correctly?

1,842 Reply by RickyAstle98 (edited by RickyAstle98 09-08-2024 20:28:26)

  • Thanks: 48
  • Thanks for the post: 1

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

dawkinscm wrote:
RickyAstle98 wrote:
dawkinscm wrote:

As he said above, on average he got around an 8 fps improvement. But this stat should be taken in the context of dev notes that suggest there might be be specific situations where the performance might be slightly worse.

My actual scenarios
v14test3 320fps / v15.2 310fps (4.4v2 with 1.78 dynamic shape and 320 fps with 2.35 dynamic shape)
Starting models 4.16 everything gets worse, just a performance drops variables from 6 to 12% respectively, the movie dynamic shapes always faster (even when the targeted resolution is slightly higher)

So it works better with Performance Boost disabled? Am I understanding this correctly?

Yes and no! Easier explanation - depending on the aspect ratio of the sources, depending on the resolution and the model used, performance can be either increased to 2% or degraded to 12% with or without TRT boost option! Seek performance is night and day with boost enabled! I did not regret that I switched to a new library!
So - with 15.2 you gets maximum 12% performance degradation, but practically is just 6% of that!

1,843 Reply by dawkinscm (edited by dawkinscm 09-08-2024 21:26:33)

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

RickyAstle98 wrote:
dawkinscm wrote:
RickyAstle98 wrote:

My actual scenarios
v14test3 320fps / v15.2 310fps (4.4v2 with 1.78 dynamic shape and 320 fps with 2.35 dynamic shape)
Starting models 4.16 everything gets worse, just a performance drops variables from 6 to 12% respectively, the movie dynamic shapes always faster (even when the targeted resolution is slightly higher)

So it works better with Performance Boost disabled? Am I understanding this correctly?

Yes and no! Easier explanation - depending on the aspect ratio of the sources, depending on the resolution and the model used, performance can be either increased to 2% or degraded to 12% with or without TRT boost option! Seek performance is night and day with boost enabled! I did not regret that I switched to a new library!
So - with 15.2 you gets maximum 12% performance degradation, but practically is just 6% of that!

OK thanks. So this is basically restating my point about @aloola's stats:

dawkinscm wrote:

..on average he got around an 8 fps improvement. But this stat should be taken in the context of dev notes that suggest there might be be specific situations where the performance might be slightly worse.

  • Thanks: 12

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

The movie Twisters (2024) is unpleasant to watch. 4K---1440p. When the characters broadcast, there are constant glitches on the screen. TRT9.2/4.15 lite. Can this be fixed? If I understand correctly, any other movie can have the same thing when the frame rate changes.

  • Thanks: 1

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

Question here.

When replacing the TensorRT model, do you use vsmlrt-windows-x64-cuda.v15.2.7z or vsmlrt-windows-x64-tensorrt.v15.2.7z?

One infers it's using Cuda while the other is actually using your RT cores? They both seem to work and don't seem to have any performance differences between the two. They also seem to perform the same as the version SVP originally came with on my 4090.

1,846 Reply by flowreen91 (edited by flowreen91 16-08-2024 12:38:49)

  • Thanks: 74

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

Bensam123 wrote:

do you use vsmlrt-windows-x64-cuda.v15.2.7z or vsmlrt-windows-x64-tensorrt.v15.2.7z?

It seems unpacking the cuda archive and deleting all the extra files doesn't make any difference on both TensorRT and ncnn/Vulkan models.
I think it's safe to use the vsmlrt-windows-x64-tensorrt.v15.2.7z instead of the uselessly big cuda archive.

Bensam123 wrote:

my 4090.

Nice! Does the 4090 haz enough power for u to watch 2x on 4K and 2K videos with latest TensorRT + latest models?

  • Thanks: 1
  • Thanks for the post: 1

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

Bensam123 wrote:

Nice! Does the 4090 haz enough power for u to watch 2x on 4K and 2K videos with latest TensorRT + latest models?

I generally watch anime at 1080p and x4 target, so somewhere around 90fps. Utilization is around 70% and can't turn it up to 5x without stuttering issues, at least for now. AIM 4.15.

I don't watch 2 or 4k content as its on a gaming PC.

1,848 Reply by Insindro

  • Thanks: 3

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

Insindro wrote:

Can anyone tell if there’s a way to add a delay to svp’s injection?

Can anyone let me know if this is possible? Thanks!

1,849 Reply by lurker

  • Thanks: 5

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

Where is the number of pre-prepared frames in SVP set? As I understand it, it generates a parameter for the vapoursynth filter from somewhere, but I did not find where this parameter is set.
My video card can barely handle 3840x1600 video. Loading is 84-100% and sometimes stutters. I think this can be solved by increasing the buffer.
https://ibb.co/37WhcJR

  • Thanks: 74

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

lurker wrote:

pre-prepared frames

concurrent-frames i think is the frc.threads from here:
https://gyazo.com/a12c49b425024e65b789ff0d4a190db8