• Thanks: 2

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

V2 won't work with me as well.

977 Reply by dawkinscm (edited by dawkinscm 13-11-2023 21:10:22)

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

Chainik wrote:

> only the folder issue is still present

no idea, it's still
> os.path.expandvars("%APPDATA%\\SVP4\\cache\\")
in vsmlrt.py

Yeah that stop working for me some time ago as well. Another issue is that I can't even find the original cache temp folder the unmodified code requests from the OS. I was able to some time ago but not now.

978 Reply by scb

  • scb
  • Member
  • Offline
  • Thanks: 8

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

Is it likely we'll see optimisations to improve performance, or is this pretty much going to stay the same even if quality improves?

  • Thanks: 86

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

V2 doesn't work for me either

rife_mnum = '4.6_v2'
trt_backend = Backend.TRT(force_fp16=True,device_id=0,num_streams=2,output_format=1,use_cuda_graph=True,workspace=None,static_shape=False,min_shapes=[64,64],opt_shapes=[1920,1088],max_shapes=[2560,1440])

        else:
                smooth = RIFE_imp(input_m,multi=rife_num,model=rife_mnum,backend=trt_backend)
               
I checked the code, "force_fp16=True" won't work with v2, should be replaced with "fp16=True"

also missing implementation = 2

I saw implementation = 2 before, but after the newest update, this was missing.

also, V1's caches stored in appdata
while V2's stored in C:\Program Files (x86)\SVP 4\rife\models\rife_v2

I freshly installed SVP and didn't modify anything.

  • Thanks: 1

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

I've tried all I could think of to make V2 models work but it won't. I've tried manually updating the TRTbackend files from the latest release package on the github. I've updated the vstrt.dll, and the files in vsmlrt-cuda with the latest version which enables TensorRT 9.1.0. The engine cache then gets built by the latest version, it's also much faster on building it compared to the current 8.5.x version.

However then both V1 and V2 models fail to work with SVP4. I've tried also updating the vsmlrt.py file with the latest one on the github, but it doesn't help.

Also both my V1 and V2 cache gets built in the C:\Program Files (x86)\SVP 4\rife\models folders and not appdata. Not like its a problem tbh, as it reads it from there also. I think this could be due to os.path not being available by the python or something.

Hopefully @Chainik can take a look and maybe also update the TRT backends, as they feature quite faster compilation of cache files.

  • Thanks: 1622
  • Thanks for the post: 1

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

aloola
> "force_fp16=True" won't work with v2, should be replaced with "fp16=True"

this is the only error I see
replace rife/helpers.py

regarding cache path - I really have no idea what's wrong and why
can't reproduce on any of my test environments

---
I can only think that you run both SVP and video player as admin, so C:\Program Files (x86) is writable in which case vsmlrt.py put cache here.
DO NOT run SVP as admin! big_smile

  • Thanks: 86

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

I still got the error after updating helpers, tried with mpv/mpc. also, everything runs without admin. here are my logs

https://cdn.discordapp.com/attachments/ … 1d44b96.py
https://cdn.discordapp.com/attachments/ … 165337.log

https://cdn.discordapp.com/attachments/290709370600423424/1173924675663966278/image.png

V1 runs fine for me.

  • Thanks: 1622
  • Thanks for the post: 1

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

this log is with an old helpers.py

984 Reply by aloola (edited by aloola 14-11-2023 11:13:34)

  • Thanks: 86

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

Chainik wrote:

This log is with an old helpers.py

OK v2 4.6 works now! thanks!

damn, I'm pretty sure I updated the file before...

also, V2 doesn't need add/crop border, I think tweaking this will give us more performance boost.

EDIT: V2 4.9 doesn't work due to not enough GPU memory, hmm weird??
https://cdn.discordapp.com/attachments/ … 175352.log

edit2: with performance boost enabled it works.
edit3: would be nice if we have an option for workspace, the default value None might cause some GPUs to run out of memory, I would like 1024, 2048, 4096 value options wink

985 Reply by donnieyeen (edited by donnieyeen 14-11-2023 11:09:44)

  • Thanks: 1

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

@Chainik @aloola did any of you try the new TensorRT 9.1.0?

v14.test2 pre-release uses trt 9.1.0 + cuda 12.2.2 + cudnn 8.9.5, which can only run on driver >= 525 and 10 series and later gpus, with improved support for self-attentions found in transformer models.

It does much faster compilation of cache. It used to work with previous version of SVP4, but now I can't get it to work anymore. I am going to buy the SVP4 licence today, really like this program and smooth playback.

When I get home I'll test the V2 models, also @aloola shoot me your discord, can add you and we can talk some video players smile

  • Thanks: 86

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

donnieyeen wrote:

@Chainik @aloola did any of you try the new TensorRT 9.1.0?

v14.test2 pre-release uses trt 9.1.0 + cuda 12.2.2 + cudnn 8.9.5, which can only run on driver >= 525 and 10 series and later gpus, with improved support for self-attentions found in transformer models.

It does much faster compilation of cache. It used to work with previous version of SVP4, but now I can't get it to work anymore. I am going to buy the SVP4 licence today, really like this program and smooth playback.

9.1 just won't work and it won't boost RIFE performance. so well let's wait for trt 10
https://github.com/AmusementClub/vs-mlrt/discussions/65

987 Reply by dawkinscm (edited by dawkinscm 14-11-2023 12:21:16)

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

donnieyeen wrote:

@Chainik @aloola did any of you try the new TensorRT 9.1.0?

v14.test2 pre-release uses trt 9.1.0 + cuda 12.2.2 + cudnn 8.9.5, which can only run on driver >= 525 and 10 series and later gpus, with improved support for self-attentions found in transformer models.

It does much faster compilation of cache. It used to work with previous version of SVP4, but now I can't get it to work anymore. I am going to buy the SVP4 licence today, really like this program and smooth playback.

When I get home I'll test the V2 models, also @aloola shoot me your discord, can add you and we can talk some video players smile

I got it working last week but it doesn't very well consumer GPUs.

  • Thanks: 1

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

aloola wrote:
donnieyeen wrote:

@Chainik @aloola did any of you try the new TensorRT 9.1.0?

v14.test2 pre-release uses trt 9.1.0 + cuda 12.2.2 + cudnn 8.9.5, which can only run on driver >= 525 and 10 series and later gpus, with improved support for self-attentions found in transformer models.

It does much faster compilation of cache. It used to work with previous version of SVP4, but now I can't get it to work anymore. I am going to buy the SVP4 licence today, really like this program and smooth playback.

9.1 just won't work and it won't boost RIFE performance. so well let's wait for trt 10
https://github.com/AmusementClub/vs-mlrt/discussions/65

I've had 9.1.0 working fine on previous version before this last update of SVP. And I was talking about the time it takes before it generated  the engine cache files, its faster then 8.6.

  • Thanks: 86

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

donnieyeen wrote:
aloola wrote:
donnieyeen wrote:

@Chainik @aloola did any of you try the new TensorRT 9.1.0?

v14.test2 pre-release uses trt 9.1.0 + cuda 12.2.2 + cudnn 8.9.5, which can only run on driver >= 525 and 10 series and later gpus, with improved support for self-attentions found in transformer models.

It does much faster compilation of cache. It used to work with previous version of SVP4, but now I can't get it to work anymore. I am going to buy the SVP4 licence today, really like this program and smooth playback.

9.1 just won't work and it won't boost RIFE performance. so well let's wait for trt 10
https://github.com/AmusementClub/vs-mlrt/discussions/65

I've had 9.1.0 working fine on previous version before this last update of SVP. And I was talking about the time it takes before it generated  the engine cache files, its faster then 8.6.

I know it works but it has a critical bug, crashed on seek or crashed when entering fullscreen. at least I tested with mpv. haven't tested with mpc yet.

  • Thanks: 1

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

The helpers.py update helped fixing the V2 models. I've tried 4.12 V2 model but its quite glitchy,.. not good to be honest, at least on my movies I tested.

I'll try 4.9 model today to see if its any better.

991 Reply by Mardon85 (edited by Mardon85 15-11-2023 18:24:25)

  • Thanks: 13

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

I can't do x3 movie speed in 4K HDR in 4.9 like I did with 4.6 is this expected behaviour?

Is there anyway to move the old cache into the new location to save caching all the resolutions again?

EDIT

Ok now I have no idea what is going on. I rebooted my machine and 4.9 has gone entirely and RIFE isn't kicking in at all??

  • Thanks: 13

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

I had to do a full reinstall and it now works.

I love how much quicker 4.9 is at compiling.
There does seem to be a bug where it doesn't like to get going after pausing which I've not seen before. It stutters down to 1FPS. I have to pause, take it back a couple of seconds then press play.

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

Mardon85 wrote:

There does seem to be a bug where it doesn't like to get going after pausing which I've not seen before. It stutters down to 1FPS. I have to pause, take it back a couple of seconds then press play.

This happened a few times while I was watching Oppenheimer yesterday when I was pausing as well. Also a weird bug where after 2 hours into the movie, it slowed down then back to around minute 50. But who knows, could have been an issue on my end maybe. Haven't tested enough to determine if it's a RIFE/svp issue yet, or something on my end.

994 Reply by aloola (edited by aloola 17-11-2023 08:23:47)

  • Thanks: 86

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

can anyone run model 4.7+ V2 with "performance boost" Disabled?

I can run model v2 4.7>4.11 with "performance boost" enabled but not when disabled

also, 4.12 lite is out.

my benchmark on blank clip 1080p, all model v1

Output 8000 frames in 34.47 seconds (232.12 fps) 4.0
Output 8000 frames in 34.14 seconds (234.35 fps) 4.4
Output 8000 frames in 33.90 seconds (235.99 fps) 4.6
Output 8000 frames in 45.65 seconds (175.26 fps) 4.7
Output 8000 frames in 46.70 seconds (171.29 fps) 4.9
Output 8000 frames in 63.21 seconds (126.56 fps) 4.11
Output 8000 frames in 63.36 seconds (126.26 fps) 4.12
Output 8000 frames in 50.69 seconds (157.83 fps) 4.12 lite


    4.0: 98.36%
    4.4: 99.30%
    4.6: 100.00%
    4.7: 74.31%
    4.9: 72.58%
    4.11: 53.64%
    4.12: 53.48%
    4.12 lite: 66.88%

995 Reply by dawkinscm (edited by dawkinscm 17-11-2023 09:12:14)

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

aloola wrote:

can anyone run model 4.7+ V2 with "performance boost" Disabled?

I can run model v2 4.7>4.11 with "performance boost" enabled but not when disabled

also, 4.12 lite is out.

my benchmark on blank clip 1080p, all model v1

Output 8000 frames in 34.47 seconds (232.12 fps) 4.0
Output 8000 frames in 34.14 seconds (234.35 fps) 4.4
Output 8000 frames in 33.90 seconds (235.99 fps) 4.6
Output 8000 frames in 45.65 seconds (175.26 fps) 4.7
Output 8000 frames in 46.70 seconds (171.29 fps) 4.9
Output 8000 frames in 63.21 seconds (126.56 fps) 4.11
Output 8000 frames in 63.36 seconds (126.26 fps) 4.12
Output 8000 frames in 50.69 seconds (157.83 fps) 4.12 lite


    4.0: 98.36%
    4.4: 99.30%
    4.6: 100.00%
    4.7: 74.31%
    4.9: 72.58%
    4.11: 53.64%
    4.12: 53.48%
    4.12 lite: 66.88%

The aim of the 4.12 Lite version is to make it less GPU intensive of 4.9. So it should arguably be called 4.9 Lite. But I'm just glad they are releasing updates and trying things.

  • Thanks: 86

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

dawkinscm wrote:
aloola wrote:

can anyone run model 4.7+ V2 with "performance boost" Disabled?

I can run model v2 4.7>4.11 with "performance boost" enabled but not when disabled

also, 4.12 lite is out.

my benchmark on blank clip 1080p, all model v1

Output 8000 frames in 34.47 seconds (232.12 fps) 4.0
Output 8000 frames in 34.14 seconds (234.35 fps) 4.4
Output 8000 frames in 33.90 seconds (235.99 fps) 4.6
Output 8000 frames in 45.65 seconds (175.26 fps) 4.7
Output 8000 frames in 46.70 seconds (171.29 fps) 4.9
Output 8000 frames in 63.21 seconds (126.56 fps) 4.11
Output 8000 frames in 63.36 seconds (126.26 fps) 4.12
Output 8000 frames in 50.69 seconds (157.83 fps) 4.12 lite


    4.0: 98.36%
    4.4: 99.30%
    4.6: 100.00%
    4.7: 74.31%
    4.9: 72.58%
    4.11: 53.64%
    4.12: 53.48%
    4.12 lite: 66.88%

The aim of the 4.12 Lite version is to make it less GPU intensive of 4.9. So it should arguably be called 4.9 Lite. But I'm just glad they are releasing updates and trying things.

have you tested with SVP? The SVP script needs to be updated to work with 4.12 lite.
you can easily check this by going to the cache folder to see if there is any model cache name 4.12_lite

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

> The SVP script needs to be updated to work with 4.12 lite.

the way vsmlrt authors work with file names is just ... ridiculous big_smile
if tomorrow there will be "4.12_super" then the script won't work again

998 Reply by dawkinscm (edited by dawkinscm 17-11-2023 11:39:37)

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

aloola wrote:
dawkinscm wrote:
aloola wrote:

can anyone run model 4.7+ V2 with "performance boost" Disabled?

I can run model v2 4.7>4.11 with "performance boost" enabled but not when disabled

also, 4.12 lite is out.

my benchmark on blank clip 1080p, all model v1

Output 8000 frames in 34.47 seconds (232.12 fps) 4.0
Output 8000 frames in 34.14 seconds (234.35 fps) 4.4
Output 8000 frames in 33.90 seconds (235.99 fps) 4.6
Output 8000 frames in 45.65 seconds (175.26 fps) 4.7
Output 8000 frames in 46.70 seconds (171.29 fps) 4.9
Output 8000 frames in 63.21 seconds (126.56 fps) 4.11
Output 8000 frames in 63.36 seconds (126.26 fps) 4.12
Output 8000 frames in 50.69 seconds (157.83 fps) 4.12 lite


    4.0: 98.36%
    4.4: 99.30%
    4.6: 100.00%
    4.7: 74.31%
    4.9: 72.58%
    4.11: 53.64%
    4.12: 53.48%
    4.12 lite: 66.88%

The aim of the 4.12 Lite version is to make it less GPU intensive of 4.9. So it should arguably be called 4.9 Lite. But I'm just glad they are releasing updates and trying things.

have you tested with SVP? The SVP script needs to be updated to work with 4.12 lite.
you can easily check this by going to the cache folder to see if there is any model cache name 4.12_lite

Yes I tested it and as discussed it is almost identical to 4.9. I saw that SVP couldn't handle the name so I just renamed it and it works fine.

@Chainik Have you had a chance to look at my mpv question?

999 Reply by lazyc97 (edited by lazyc97 17-11-2023 12:31:20)

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

Do we have the changelog of the models somewhere?
I can only find download links on vs-mlrt and practical-RIFE repo but nothing about what's new in a version.

1,000 Reply by Drakko01 (edited by Drakko01 17-11-2023 17:15:06)

  • Thanks: 49
  • Thanks for the post: 1

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

For those who have more technical knowledge, this it's good for us, right?
https://www.google.com/amp/s/www.techpo … face%3famp