Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion.

Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules.

 

Go Back   Doom9's Forum > Capturing and Editing Video > VapourSynth

Reply
 
Thread Tools Search this Thread Display Modes
Old 20th November 2021, 20:42   #61  |  Link
lansing
Registered User
 
Join Date: Sep 2006
Posts: 1,657
I don't understand, what was the filter supposed to do??
lansing is offline   Reply With Quote
Old 20th November 2021, 21:01   #62  |  Link
Selur
Registered User
 
Selur's Avatar
 
Join Date: Oct 2001
Location: Germany
Posts: 7,313
I'm not sure either which is why is simply took a clip and made some screen shots and asked whether someone has experience with it,..from what I have seen so far each of the models has a different effect.
So far I would say:
deblur GoPro: seems to mainly sharpen and 'add details' (seems to causes artifacts on cartoons)
deblur REDS: strop denosing and sharpening (seems to only suited for cartoon content)
denoise: also seems to be useful for really noise vhs sources or cartoons
derain: milder denoise/degrain, unsure where this could be useful, seems to remove less when used on more detailed content,...
-> so no real clue
__________________
Hybrid here in the forum, homepage

Last edited by Selur; 20th November 2021 at 22:28.
Selur is offline   Reply With Quote
Old 20th November 2021, 21:22   #63  |  Link
poisondeathray
Registered User
 
Join Date: Sep 2007
Posts: 5,390
for hinet: the main purpose is single image deblurring . It's not a temporal algorithm. The type of blur it's most suited to - is mostly camera motion blur / shutter blur.

"derain" is for removing "rain" (it's not "degrain") . ie. short streaks of white, like rain drops. It actually works ok on random dark film scratches (not persistent , long scratches) if you invert the src.

Another semi popular deblurring algorithm is deblurGANv2 (also works best on camera motion blur)
poisondeathray is offline   Reply With Quote
Old 20th November 2021, 23:45   #64  |  Link
lansing
Registered User
 
Join Date: Sep 2006
Posts: 1,657
Quote:
Originally Posted by poisondeathray View Post
for hinet: the main purpose is single image deblurring . It's not a temporal algorithm. The type of blur it's most suited to - is mostly camera motion blur / shutter blur.

"derain" is for removing "rain" (it's not "degrain") . ie. short streaks of white, like rain drops. It actually works ok on random dark film scratches (not persistent , long scratches) if you invert the src.

Another semi popular deblurring algorithm is deblurGANv2 (also works best on camera motion blur)
Ok thanks. So it's targeting high resolution photo images, not videos.
lansing is offline   Reply With Quote
Old 21st November 2021, 00:23   #65  |  Link
poisondeathray
Registered User
 
Join Date: Sep 2007
Posts: 5,390
Quote:
Originally Posted by lansing View Post
Ok thanks. So it's targeting high resolution photo images, not videos.
Not necessarily; the gopro model is trained on gopro video . The resolution isn't "high", though

All the single image algorithms tend to have some temporal inconsistencies in the end result when used on video .

YMMV for all these research projects
poisondeathray is offline   Reply With Quote
Old 5th December 2021, 08:49   #67  |  Link
kedautinh12
Registered User
 
Join Date: Jan 2018
Posts: 2,160
Wow, interesting results
kedautinh12 is offline   Reply With Quote
Old 5th December 2021, 09:58   #68  |  Link
Selur
Registered User
 
Selur's Avatar
 
Join Date: Oct 2001
Location: Germany
Posts: 7,313
Doesn't really look better than BasicVSR++ to me atm. but in case someone ports it to Vapoursynth we will see more.
__________________
Hybrid here in the forum, homepage
Selur is offline   Reply With Quote
Old 5th December 2021, 17:41   #69  |  Link
Gravitator
Registered User
 
Join Date: May 2014
Posts: 292
At the end, stripes appear under the left arm in Titanic.
__________________
Win10x64, Xeon E5450, GTX 750 2GB, DDR3 8GB.
Gravitator is offline   Reply With Quote
Old 23rd December 2021, 09:17   #70  |  Link
PatchWorKs
Registered User
 
PatchWorKs's Avatar
 
Join Date: Aug 2002
Location: Italy
Posts: 305
Quote:
Originally Posted by Selur View Post
Doesn't really look better than BasicVSR++ to me atm. but in case someone ports it to Vapoursynth we will see more.
Some other "real world" comparisons done with Colab Demo:
PatchWorKs is offline   Reply With Quote
Old 28th December 2021, 22:25   #71  |  Link
Selur
Registered User
 
Selur's Avatar
 
Join Date: Oct 2001
Location: Germany
Posts: 7,313
Here's a nice example what BasicVSR++ can do:


Cu Selur
__________________
Hybrid here in the forum, homepage
Selur is offline   Reply With Quote
Old 31st December 2021, 05:44   #72  |  Link
aegisofrime
Registered User
 
Join Date: Apr 2009
Posts: 478
Can I get a bit of help here? After finally setting up the dependencies (they are a bit convoluted for this plugin honestly), I keep getting the following error:

Code:
vpy  [error]: CUDA out of memory. Tried to allocate 1.24 GiB (GPU 0; 8.00 GiB total capacity; 5.08 GiB already allocated; 0 bytes free; 5.21 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation occurred while getting frame 0
x265 [error]: unable to open input file <P:\ISO\MainMovie\ME_021\OME021_temp\OME021.vpy>
My Script is very simple,

Code:
clip = core.avs.DGSource(r"D:\ISO\Test.dgi")
#core.max_cache_size=16384
#core.num_threads = 8
import havsfunc as haf
core.max_cache_size=16384
clip = haf.QTGMC(clip, Preset='Placebo', TFF=True, Denoiser="KNLMeansCL")
from vsbasicvsrpp import BasicVSRPP
clip = core.resize.Bicubic(clip=clip, format=vs.RGBS, matrix_in_s="470bg", range_s="limited")
clip = BasicVSRPP(clip,model=1)
clip = core.resize.Bicubic(clip=clip, format=vs.YUV420P12, matrix_s="470bg", range_s="limited")
clip = core.resize.Spline36(clip, 1440,960)
clip.set_output()
Input is just a 720x480 DVD video, while my GPU is a RTX 3060Ti, so it only has got 8GB of RAM. It should still be enough right? How can I tweak the arguments so that this works at all?
aegisofrime is offline   Reply With Quote
Old 31st December 2021, 08:37   #73  |  Link
Selur
Registered User
 
Selur's Avatar
 
Join Date: Oct 2001
Location: Germany
Posts: 7,313
adding:
Code:
,fp16=True
and/or
Code:
,tile_x=360,tile_y=480
to the BasicVSRPP code should help.

Cu Selur
__________________
Hybrid here in the forum, homepage
Selur is offline   Reply With Quote
Old 2nd January 2022, 09:33   #74  |  Link
PatchWorKs
Registered User
 
PatchWorKs's Avatar
 
Join Date: Aug 2002
Location: Italy
Posts: 305
For those - like me - who don't own powerful GPUs, I just asked kodxana (aka Mediator2011):

https://github.com/kodxana/VapourSynthColab/issues/1
PatchWorKs is offline   Reply With Quote
Old 18th January 2022, 23:03   #75  |  Link
PatchWorKs
Registered User
 
PatchWorKs's Avatar
 
Join Date: Aug 2002
Location: Italy
Posts: 305
Hi there, can someone make some (performances, particularry) tests of VapourSynth Single Image Super-Resolution Generative Adversarial Network (GAN) ?
(note: I tried their colab without success)

Thanks in advance !

Last edited by PatchWorKs; 18th January 2022 at 23:16.
PatchWorKs is offline   Reply With Quote
Old 20th January 2022, 05:58   #76  |  Link
Selur
Registered User
 
Selur's Avatar
 
Join Date: Oct 2001
Location: Germany
Posts: 7,313
What do you have in mind?
__________________
Hybrid here in the forum, homepage
Selur is offline   Reply With Quote
Old 20th January 2022, 18:44   #77  |  Link
PatchWorKs
Registered User
 
PatchWorKs's Avatar
 
Join Date: Aug 2002
Location: Italy
Posts: 305
Quote:
Originally Posted by Selur View Post
What do you have in mind?
Oh, well, it would be great to have a colab "service" where upload (old) SD stuff and download (x265-10bit encoded) 1080p...
PatchWorKs is offline   Reply With Quote
Old 20th January 2022, 18:56   #78  |  Link
Selur
Registered User
 
Selur's Avatar
 
Join Date: Oct 2001
Location: Germany
Posts: 7,313
Ah okay, sounded like wanted some benachmarks. No clue about colab.
__________________
Hybrid here in the forum, homepage
Selur is offline   Reply With Quote
Old 2nd February 2022, 17:32   #79  |  Link
PatchWorKs
Registered User
 
PatchWorKs's Avatar
 
Join Date: Aug 2002
Location: Italy
Posts: 305
Quote:
Originally Posted by Selur View Post
Ah okay, sounded like wanted some benachmarks. No clue about colab.
Oh, well, of course it would be great to have some tests of newest algos:

https://github.com/Weepingchestnut/A...per-Resolution
PatchWorKs is offline   Reply With Quote
Old 2nd February 2022, 18:02   #80  |  Link
poisondeathray
Registered User
 
Join Date: Sep 2007
Posts: 5,390
PP-MSVSR featured in PaddleGAN looks like the next improvement over basicvsr++ . I'm trying to get it to work, but no luck so far with modifying config files and paths

https://github.com/PaddlePaddle/Padd..._resolution.md
https://arxiv.org/pdf/2112.02828.pdf
poisondeathray is offline   Reply With Quote
Reply

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +1. The time now is 16:21.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2024, vBulletin Solutions Inc.