Search options (Page 5 of 6)

On tensor rife 4.6 %15 scene change threshold i get lag on fast panning scenes just like on vulkan 4.4 with %10 scene change threshold. My gpu usage is around %50-55. Is there a solution for this ?

Chainik, please try implement new ai scene detection from https://github.com/styler00dollar/VSGAN … detection. I use this fragment from Mad Max for tests how scene detection algorithms working https://drive.google.com/file/d/1f7zZ_A … share_link and this new thing performed almost perfectly.

SVP updated, added TensorRT (no need for doing anything manually), fixed (??) possible audio desync after RIFE transcoding (both Vulkan and TRT versions)

Either select the checkbox at install time, or install an additional package called "TensorRT" via SVP menu -> Utilities -> Additional programs and features.

----
It uses "dynamic shapes" by default, i.e. only ONE optimization pass for all resolutions below 2560*1440. And if/when you open a 4K video then it'll do another pass for resolutions from 1440p to 4K.
BUT! this can be a little bit slower than optimization for a specific resolution, like 5% in my case. This is why there's a "performance boost" profile option - it forces its own opt. pass for each resolution.

another finding: model 4.6 needs more than 8G of device memory for 4K res, so if you're on a 8G card and want 4K then choose model 4.4.

----
Also installation scripts fixed so the Vapoursynth Filter must be registered automatically.

edit this line

clip = RIFE(clip=clip, multi=interpMulti, model=RIFEModel.v4_6, backend=Backend.TRT(fp16=True, device_id=GPU, workspace=WS_size, use_cuda_graph=True, num_streams=GPU_t))

I think 2000 is enough for 4k

Chainik wrote:

>
dunno, I don't see any real problem with the current simple and fast method
another neural network just for a scene change detection looks like an overkill to me

Just try transcoding that clip from Mad Max. SCDetection works very bad. To avoid freezes, you need set more than 15%, I set 20%. At 15%, there are still freezes. But at high values,  ends of scenes are no longer detected at all and they simply flow one into another. It looks bad. I just want quality, not just work. I think I'm not the only one. By the way, when I tested, the difference with and without this system was no more than 10%. Fast scenes go with fading, part of scenes flow into one another, this is all work that does SCDetect.

I deleted SVP and startet from scratch, just to check if the inscructions are complete and everything is working.
There is a step missing.
After replacing generate.js and base.py, start SVP4, add the new option TensorRT etc.

Then the missing step:
Copy the Rife AI profile and select the AI Model "rife"
Enable the new Option TensorRT On

updated instructions

should improve FPS on 4080-and-better (probably 4070 too, dunno), when performance is bound by the system's RAM bandwidth, not GPU power
i.e. for 4K playback

Depends on your resolution .. 1080p runs without any problem .. 2k as well ... 4k is still not fast enough

=== RIFE / PyTorch+TensorRT installation ===

!!! UPDATED 07-01-2023 !!!
- replaced base.py : fp16 isntead of fp32
+ added fix for vsrife
============

!!! Advanced users only, if you don't know what this is all about - you don't need it !!!

0. Update SVP via online updater (this will update Python to ver. 3.10.9)

1. Dependencies installation:
--------------
Download TensorRT libs: https://github.com/HolyWu/vs-rife/relea … 2_win64.7z
unpack CUDA-blah-blah.7z to SVP 4\mpv64

run cmd as Administrator

cd SVP 4\mpv64

python get-pip.py

python -m pip install torch torchvision --extra-index-url https://download.pytorch.org/whl/cu117
python -m pip install tensorrt-8.5.2.2-cp310-none-win_amd64.whl
python -m pip install torch-tensorrt-fx-only tqdm
python -m pip install -U --no-deps vsrife
python -m vsrife

A quick fix for vsrife's "slowmo" bug: replace __init__.py in SVP 4\mpv64\Lib\site-packages\vsrife\

2. SVP modification
---------------
Replace SVP 4\script\generate.js and base.py with the attached ones

menu -> Applications settings -> Additional options -> All settings, go to 'User defined options'
Title: TensortRT; Script name: rife_trt; Other values by default - "FRC profile", "ON or OFF"
Click 'Add option'

Go to the RIFE video profile, see the added TensortRT on/off swtich.
Make a copy of the RIFE video profile, one with TensorRT ON and another with OFF.

----------------
For a given video resolution the first-time start is super-slow (while it generates TRT net in a cache folder --> %APPDATA%\SVP4\cache\trt), so better do it in a "transcoding" mode.

Now I am using a project named mpv_lazy in github(https://github.com/hooke007/MPV_lazy, in Chinese), and it enables RFIE v4.6 using tensorrt. My video card is 3070 laptop and it works totally fine in 1080p real-time doubling FPS, the performance of which is much better than the SVP4. Hope the svp4 can support the trt RFIE as soon as possble.