Posts: 27
	Threads: 1
	Joined: Nov 2022
	
	
 
	
	
		Dear Selur.
I have three GPUs - two TITAN RTX and one 4090RTX.
With version Hybrid_2023.03.17.1 and VapoursynthR61_torch_2023.03.12
BasicVSR++ runs in three threads. With version Hybrid_dev_2023.04.18-205712
no way.
	
	
	
	
	
 
 
	
	
	
		
	Posts: 12.056
	Threads: 66
	Joined: May 2017
	
	
 
	
	
		Still even with 24GB RAM  on the 4090 RTX, when filtering 4k content with BasicVSR++ you will run out of vram unless you use tiling. 

Does the Vapoursynth Preview show an error when you open it?
If it just closes, save the script to a file and see if you can open it without issues with vsViewer.
=> Simply complaining that is does not work will not allow me to figure out what the issue might be.
Cu Selur
	
 
	
	
----
Dev versions are in the 'experimental'-folder of my GoogleDrive, which is linked on the download page.
	
	
 
 
	
	
	
		
	Posts: 12.056
	Threads: 66
	Joined: May 2017
	
	
 
	
		
		
		19.04.2023, 18:44 
(This post was last modified: 19.04.2023, 18:45 by Selur.)
		
	 
	
		Ah, just noticed current torch is 2023.04.16 

-> will send the link
Cu Selur
	
 
	
	
----
Dev versions are in the 'experimental'-folder of my GoogleDrive, which is linked on the download page.
	
	
 
 
	
	
	
		
	Posts: 987
	Threads: 81
	Joined: Feb 2020
	
	
 
	
	
		Hello Selur,
  your dev version is working perfectly. With this update using BasicVSR++ with the LinearMask (that allows to use weighted merge) transform BasicVSR++ in one of the best VSR filters available, especially with model NTIRE 2021 (2), which is more aggressive (so the mask is necessary) in introducing more details in the video.
Thanks,
Dan
	
	
	
	
	
 
 
	
	
	
		
	Posts: 12.056
	Threads: 66
	Joined: May 2017
	
	
 
	
	
		Happy, it's working in the way you wanted it too.
	
	
	
----
Dev versions are in the 'experimental'-folder of my GoogleDrive, which is linked on the download page.
	
	
 
 
	
	
	
		
	Posts: 27
	Threads: 1
	Joined: Nov 2022
	
	
 
	
	
		Dear Selur.
Everything worked with VapoursynthR61_torch_2023.04.16.
Regarding LinearMask, I completely agree with Dan64-
additional filter adjustment appears.
Thank you.
	
	
	
	
	
 
 
	
	
	
		
	Posts: 12.056
	Threads: 66
	Joined: May 2017
	
	
 
	
	
		As a "head's up": I'm going to remove the 'LinearMask' and replace it with a general 'Merge'-option.
http://www.vapoursynth.com/doc/functions...merge.html
LinearMask was a special case of Merge:
core.std.Merge(clipa=filteredClip, clipb=unfilteredClip, weight=[setWeight])
So, instead of enabling "Show 'Masked' -controls" you will have to enable "Show 'Merge'-controls" and set the weights of each plane to the weight you want.
Cu Selur
	
 
	
	
----
Dev versions are in the 'experimental'-folder of my GoogleDrive, which is linked on the download page.
	
	
 
 
	
	
	
		
	Posts: 987
	Threads: 81
	Joined: Feb 2020
	
	
 
	
	
		Hello Selur,
   thanks for the update. It will be possible to automatically set all the planes to the same value ?
   Just for curiosity, did you see any advantage in assigning different values to the planes ? 
Dan
	
	
	
	
	
 
 
	
	
	
		
	Posts: 12.056
	Threads: 66
	Joined: May 2017
	
	
 
	
	
		Atm. I haven't planned to add additional io elements to set all values at the same time.
The general idea is to allow (not for just BasicVSR++) to allow to just for example use a specific plane, this basically adds to all filters the option to restrict them to specific planes.
	
	
	
----
Dev versions are in the 'experimental'-folder of my GoogleDrive, which is linked on the download page.