You are not logged in. Please login or register.
SmoothVideo Project → Posts by dawkinscm
What is the difference between 4.9 and 4.12 lite versions ?
If I would guess lower picture quality better performance ?
And if so is it easier on CPU or GPU or Both ?
Very little difference. So try both and let us know what you think.
I think your post will be reverenced many times going into the future...
Everything needed is already on this SVP manuals page https://www.svp-team.com/wiki/RIFE_AI_interpolation The extra stuff you've seen some of us doing here is interesting but not necessary for Rife. Especially since the SVP devs recently created a script that does most of the extra stuff for you.
The instructions you need are as follows:
1. Download and install SVP.
2. Follow the Rife AI manuals page.
3. Restart SVP.
4. Don't worry about the experimental stuff myself and others are doing because for most it's more likely to break your install than help it.
don't you know that SVP works via vapoursynth filter?
I'm pretty good at getting things working without know too much about them then I eventually learn more as I go along. I am still learning about all of this stuff. I know nothing about vapoursynth except that it is similar to avisynth and that has it's own library of ffmpeg based filters?. So since mpv uses filters based on ffmpeg I assume that vapoursynth is using the same filters?
I checked out base.avs but saw nothing there.
update now
Sorry to bother you but as I asked for above, where would I look in future within SVP code for something like this? Where were/are the mpv filters used?
Thanks. I would never have guess that SVP would be the cause of this. For future reference, where in SVP does it use/remove mpv filters?
The changelog says:
= adjusted for recent mpv (pre-0.37) changes
Does that include making sure that deprecated options have been accounted for?
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 lite4.0: 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?
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 lite4.0: 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.
I'm getting an error message when SVP stops working properly with the latest versions of mpv and Plex. This doesn't happen on the first video played. It happens after the first video is stopped or finished. The once the pipe/socket is disconnected it doesn't seem to get reconnected to any other video until I exit Plex and start again. The error message says "Option vf-del : 0 Doesn't exist" because they the "-del" option has been deprecated in the latest version.
@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
I got it working last week but it doesn't very well consumer GPUs.
> 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.
Which model is currently the best for non-anime content? Most discussions I see are based around anime.
Also, what exactly do the ensemble models do differently?
Don't bother with the ensemble models. Whatever they are supposed to do isn't worth it. Also don't worry about the "anime" tag because they all work fine with non anime stuff. If you have a 30 series Nvidia GPU or above and use TensorRT then try Rife 4.9 if your GPU can handle it or Rife 4.6 and below if it can't.
SVP updated, there're a few changes regarding RIFE/trt:
- added 'akarin' plugin, which should improve seeking performance with v1 models up to v2 level
- put any models (including future ones) into SVP 4\rife\models\rife (for V1 models) and SVP 4\rife\models\rife_v2 (for V2 models) and restart SVP - they all will be listed in the video profile; "ensemble' models are supported too.
- a quick fix for DolbyVision playback (color blinking issue)
- "opt shapes" were not set with latest vsmlrt.py which resulted in engine rebuilding for every video resolution regardless of "perf. boost" option
- included model 4.9
Excellent! The new helper script detect any model you add without having to make manual changes, although v12 currently doesn't work. Rife v2 seeking was almost immediate, with akarin the seeking is immediate.
Great update! Nice work and thank you 
Edit: They've literally just released an update to v12 and it is working. Will update with results.
Results: Rife 4.12 almost identical to 4.11 but has some fixes. Rife 4.9 still the best overall for me.
Rife 4.11 is out. The file is much smaller than any other Rife model but is just as computationally intensive for GPUs and it has a couple of other small issues. Also I checked 4.9 with my hard coded subtitled files but I didn't have any warping issues so I'm staying with that.
subtitles issue you mentioned
I never mentioned this.
Subtitles was someone else.
I don't know why you try to use the newest model, just pick the one you like the most. newer doesn't mean better.
The reason new models are being released is because they are still evolving through training, so what you are suggesting makes no sense. If 4.6 works for you then fine. But the one I like the most is the one with the least artefacts and best motion and that is currently 4.9+.
...
also this is just my personal benchmark, not related to your issueall v2 1080@30 x2
4.10: 113
4.9: 145
4.8: 148
4.7: 149
4.6: 195well, 4.10 is super heavy, good for ppl with HE graphics and want quality.
Your figures confirm my findings about GPU usage. Rife 4.10 is by far the worst. There is very little quality difference between 4.9 and 4.10 but there is a massive hit on GPU performance. I will probably check out the subtitles issue you mentioned because that would be a major problem.
The recent Rife networks are being specifically trained for Anime so I suppose any improvements in live video processing is a bonus. But there are so many Anime models I'm not sure why we need another one
Rife 4.10 is out but I strongly recommend leaving it alone because of excessively high GPU usage. But if you wish to try it then YMMV.
Update: I tried v2. After reboot it only made a 1-2% difference for me.
It's been a while since I properly looked at the Python wrapper code. Looking at your post reminded me of how I handled the v2 code previously and not the "hack" I did this time.
I also added the lines below to the Python script (replacing the temp dir conditional) but it didn't work.
You can update `vsmlrt.py` by getting a copy from the `AmusementClub/vs-mlrt` repo, then modifying:
alter_engine_path = os.path.join( os.path.expandvars("%APPDATA%\\SVP4\\cache\\"), os.path.splitdrive(engine_path)[1][1:] )
I had this issue before but I can't remember how I fixed it. But since it makes no obvious difference to how Rife works I'll probably just leave it as it is for now.
it didn't change anything - i can only do rife 49 with avisynth... vapoursynth doesn't work and keeps spitting out errors.
i dont know python, but this is the error message i'm getting when using Vapoursyth:
[11/05/2023-17:16:04] [E] Error[4]: [network.cpp::nvinfer1::Network::validate::3100] Error Code 4: Internal Error (input: for dimension number 1 in profile 0 does not match network definition (got min=11, opt=11, max=11), expected min=opt=max=7).)
[11/05/2023-17:16:04] [E] Error[2]: [builder.cpp::nvinfer1::builder::Builder::buildSerializedNetwork::751] Error Code 2: Internal Error (Assertion engine != nullptr failed. )
[11/05/2023-17:16:04] [E] Engine could not be created from network
[11/05/2023-17:16:04] [E] Building engine failed
[11/05/2023-17:16:04] [E] Failed to create engine from model or file.
[11/05/2023-17:16:04] [E] Engine set up failed
Check that you don't have your own local copies of vapoursynth or avisynth installed. That will cause problems.
alter_engine_path = os.path.join( os.path.expandvars("%APPDATA%\\SVP4\\cache\\"), os.path.splitdrive(engine_path)[1][1:] )
I always forget about this. It doesn't really matter because Rife still works without it. But it is neater and I might add it back for any future update.
dawkinscm wrote:The simplest way to install a new Rife model is to copy the Rife onnx file to the SVP rife directory then modify the rife_trt_model property under the video profiles options settings in SVP.
I decided to compare Rife 4.9 v1 with Rife 4.9 v2. The good news is that as expected seek speed is almost instantaneous seeking and for me it uses about 4 percent less GPU than v1. However, even though it technically uses a little less GPU power, my GPU temps went up about 4 degrees. BTW I don't watch 4K with Rife, but so far I haven't seen any resolution issues with v2.
Could you be kind and specify where could I get 4.9 v2, please ?
https://github.com/AmusementClub/vs-mlr … e_v2_v2.7z
This is the specific URL but the page itself has lots of other models too
Today I decided to compare Rife 4.6 v2 with 4.9 v2. After a full reboot, Rife 4.9 v2 GPU usage dips below 70% into the high 60s and averages around 74%. This is better than I reported yesterday. But Rife 4.6 v2 is below 50% Both models use about 8.6GB of RAM. Please note that my GPU stats are always a little higher than most because the GPU is doing multiple things.
I get the following message on the dos window when I used 4.9 for the first time. I'm driving a RTX 3080 "Your ONNX model has been generated with INT64 weights, while TensorRT does not natively support INT64. Attempting to cast down to INT32."
We all get that. It's a warning, not an error. It's mostly fine.
The simplest way to install a new Rife model is to copy the Rife onnx file to the SVP rife directory then modify the rife_trt_model property under the video profiles options settings in SVP.
I decided to compare Rife 4.9 v1 with Rife 4.9 v2. The good news is that as expected seek speed is almost instantaneous seeking and for me it uses about 4 percent less GPU than v1. However, even though it technically uses a little less GPU power, my GPU temps went up about 4 degrees. BTW I don't watch 4K with Rife, but so far I haven't seen any resolution issues with v2.
Rife 4.9 is out. First I've got "some" GPU peformance back. Before I was peaking at 83% now I'm peaking at 79%. One of my test files no longer shows any artefacts. Another test file shows improvement and the artefact can be mitigated by reducing SC. One related artefact is still present but seems to have slightly improved as well.
s Rife 4.9 better or improved for movies in general or only focused on anime, thanks in advance.
They said Rife 4.7 was focused on Anime but it also reduced or removed artefacts from my test movies. Rife 4.9 improves upon 4.7 by removing or reducing more artefacts.
SmoothVideo Project → Posts by dawkinscm
Powered by PunBB, supported by Informer Technologies, Inc.
